WorksPhysics and Astronomy
Pentagon Functions for Two-Loop Leading-Colour Amplitudes with a Light-Quark Loop
Abstract
We present a basis of special functions and an efficient numerical implementation for the planar two-loop master integrals that enter the leading-colour amplitudes for with a closed light-quark loop. We provide analytic results for the special functions at transcendental weight one and two in terms of logarithms and dilogarithms, while higher weights are evaluated as one-fold integrals. We focus on the nested roots that arise, studying their analytic structure on the physical region and constructing numerically stable representations of letters and forms that involve them. We find that a number of forms have an antiderivative that cannot be represented by a single logarithm over the physical region and tackle them numerically with adaptive Chebyshev approximation. We implement the functions in C++ as an extension of the PentagonFunctions++ library. We study the performance of our implementation over physical phase-space points and find that the evaluation is accurate and fast enough for phenomenological application. This provides an important building block for NNLO studies of production at the LHC.
Introduction
One of the main goals of the high-luminosity physics programme of the LHC is the study of the properties of the Higgs boson. With the increased dataset that it will provide, we are expecting first meaningful constraints on the Higgs boson self-coupling, together with an increased precision of the measurements of its coupling with the other Standard Model particles. The top-quark Yukawa coupling is the largest one of them, and can be directly accessed at the LHC through the associated production of a top-quark pair and a Higgs boson, . Since its observation at the LHC [1, 2], measurements of this process have constrained deviations from the Standard-Model Yukawa coupling at the level of ten percent [3, 4], and by the end of the high-luminosity run they are expected to reach the level of a few percent, at which point theoretical uncertainties will dominate [5, 6]. This makes it essential to include next-to-next-to-leading-order (NNLO) QCD predictions in phenomenological studies. Predictions at this level of precision are by now available, where the so far missing two-loop amplitudes have been approximated in different kinematic limits, with the resulting uncertainty on the predictions estimated at the percent level. This was done first for the inclusive cross section [7, 8], where the two-loop amplitude was approximated by an expansion in the soft-Higgs limit, then for differential distributions [9], where a high-energy expansion in the small top-quark mass limit was also used. Recently, such fixed-order results have also been matched to resummation [10] and parton showers [11]. Further developments in the calculation of the amplitudes include an approximate computation in the boosted limit [12], and their exact numerical evaluation in the quark-initiated part [13] and in the leading-colour approximation for the channel [14]. Removing the need for these approximations at NNLO calls for two-loop master integrals that can be evaluated quickly and reliably everywhere in physical phase space. In this paper we provide such an implementation for the planar master integrals entering the leading-colour two-loop amplitudes with a closed light-quark loop.
These master integrals were first presented in ref. [15], representing the first two-loop five-point integral computation in a problem with seven kinematic scales and massive internal lines. There, a basis of master integrals was constructed that satisfies a system of differential equations in -factorised form. The master integrals were then evaluated numerically via generalised series expansions [16] as implemented in DiffExp [17], with boundary values from the auxiliary mass flow method [18, 19, 20, 21]. This yielded high-precision benchmarks, but evaluating the integrals at the large number of phase-space points required by an NNLO calculation remained challenging. An approach to overcome this challenge is the construction of so-called “pentagon functions” [22, 23, 24]. One identifies, at each transcendental weight, a minimal set of algebraically independent special functions in terms of which all master integrals can be expressed. Functions of weights one and two are written in terms of logarithms and dilogarithms, while functions of weights three and four are represented as one-fold integrals along a straight-line path in phase space [22]. Implemented in the PentagonFunctions++ library [24, 25, 26], this approach provides the two-loop special functions underlying many of the available NNLO QCD predictions for processes (see ref. [27] for a recent overview). Alternative strategies for evaluating multi-scale integrals include generalised series expansions [16, 17, 20, 21, 28, 29], Chebyshev approximations [30, 31], numerical differential equation solving [32, 33, 34] and finite-element approaches [35]. In this work, we follow the one-fold integral approach, which exploits directly the compact -factorised differential equation and the analytic form of the functions at low weight.
Applying the pentagon-function approach to the integrals raises new issues, which stem from the presence of nested square roots in the alphabet (a feature that has been observed since then in other complicated two-loop integrals [36, 37, 38, 39, 40]). The nested roots are non-analytic on a hyperplane in the physical phase space, even though this hyperplane is not a Landau singularity. We avoid the difficulties associated with crossing this hyperplane by placing the boundary point of our integration paths on the hyperplane. Furthermore, straightforward representations of the letters that involve nested roots, and of their forms, develop spurious singularities, which we remove by suitable rewriting. Finally, the forms of some letters have antiderivatives that cannot be written as a single logarithm over the whole phase space. In order to avoid partitioning phase space, we construct these antiderivatives numerically along the integration path with adaptive Chebyshev approximations [30].
We implement the evaluation of our set of special functions in C++, as an extension of PentagonFunctions++ [24, 25, 26], which can be downloaded from the public repository:
https://gitlab.com/pentagon-functions/PentagonFunctions-cpp
Comparing evaluations in double and quadruple precision over a sample of physical phase-space points, we find that the numerical accuracy is under excellent control, and the average evaluation of the full set of functions takes around half a second, making the functions suitable for phenomenological applications. As the same integrals appear in two-loop amplitudes for processes such as , our results are also relevant beyond production.
Our work is organised as follows. In § 2 we set up the kinematics and the square roots of the alphabet. In § 3 we recall the integral families and their differential equations. In § 4 we analyse the nested square root on physical phase space and its consequences for the integration paths, and construct the basis of pentagon functions weight by weight, giving analytic expressions at weights one and two and one-fold integral representations at weights three and four. In § 5 we describe the numerical implementation, including the treatment of the nested-root letters and the Chebyshev approximation of d log form antiderivatives, and assess its precision and performance. We conclude in § 6.
Scattering Kinematics and Notation
In this work, we consider the production of a top-quark pair in association with a Higgs boson at hadron colliders. Specifically, we work in the kinematic channel
where labels a generic initial-state massless parton. In particular, we work with on-shell top quarks and the on-shell conditions for the momenta take the form
We keep the virtuality of generic, so that the results also apply to other massive final-state bosons, such as the boson in on-shell production. By convention, we take all external momenta to be incoming, so that momentum conservation reads
The scalar Feynman integrals that we consider are functions of the Mandelstam invariants
where we define
All other invariants , such as , , , and appearing below, are linear combinations of those in .
In the phase-space channel defined by eq. (1), the kinematic invariants respect the following set of inequalities:
where is the Gram matrix. Beyond these constraints, we will also impose the additional constraint that the final-state off-shellness is less than 4 times the square of the top mass, i.e.,
Importantly, eq. (7) is fulfilled by on-shell kinematics and for most phenomenological applications for production, as well as for on-shell production.
The Feynman integrals under consideration have a number of algebraic branch points. Many of them are familiar from one-loop calculations. A number can be identified as the zeros of the following Gram determinants
Further algebraic branch points occur at zeros of the so-called “Cayley determinants”
The kinematic constraint that we imposed in eq. (7) has been chosen such that we can assume that we are always on one side of the branch point at . In particular, , so is purely imaginary throughout the region we consider. A number of the algebraic branch points of the Feynman integrals do not arise at one loop, denoted as in ref. [15]. We refer to ref. [15] for their explicit expressions. Beyond this, in ref. [15], in order to rotate to a basis of pure integrals, it was necessary to introduce a square root of a quantity which already contained a root, , defined through
As discussed in ref. [15], the square roots of and are not algebraically independent, but are related through their product, (see eq. (43)), where is times a perfect square. In order to simplify the algebraic branch structure of the pure basis, this led to the definition of an algebraic square root as
We stress that this is a definition of that is distinct from the principal-branch one: it ties the branch-cut structure of to that of . In contrast, is evaluated using the standard principal branch prescription. As we will see in § 4.3, these nested roots have a highly non-trivial structure on physical phase space.
Integral Families and Differential Equations
Integral Families
In ref. [15], it was shown that the leading-colour fermionic contribution to the two-loop amplitude can be expressed in terms of a basis of independent master integrals that is closed under the map
While the one-loop master integrals can all be expressed in terms of the pentagon integral family , the two-loop master integrals can be expressed in terms of two integral families: the penta-box and the penta-bubble integral families. We label the complete set of master integrals as , where they are listed according to the ordering associated to each topology: , , , . Within a topology, the master integrals are ordered according to the conventions fixed in [15], to which we refer the reader for their complete list.
The penta-box integral family, labelled with , refers to the topology of the left diagram in Figure 1. It is defined by 8 inverse propagators, with three additional inverse propagators introduced to complete the integral family, making it possible to express any scalar product between the loop momenta , and the external momenta in terms of inverse propagators. A generic element of the integral family is then identified by the vector of exponents of the propagators, , and can be written as:
Figure 1. The topologies of the two integral families entering our computation: the penta-box on the left and the penta-bubble on the right. (Diagrams reproduced from ref. [15].)
where , is the dimensional-regularisation scale, and propagators carry the Feynman prescription . They are defined as
We note that the integral family is invariant under the action of the transformation of eq. (20). The dimension of the vector space spanned by the integral family , and thus the number of master integrals with this topological structure, is 111.
The penta-bubble integral family refers instead to the topology on the right of Figure 1. We label it as and we write a generic element of this integral family as:
In this case, there are 6 inverse propagators directly coming from the topological structure of the diagram, and we introduce 5 additional ones to complete the integral family. We define them as
The dimension of the vector space spanned by is 19. Nevertheless, of the 19 resulting master integrals, 11 are shared with the basis of , leaving us with only 8 new master integrals. In contrast with the case of , we observe that is not invariant under the mapping given by the transformation . As a consequence, we need to consider also the vector space spanned by which thus adds 8 new master integrals to our analysis.
When computing the finite remainder for the two-loop light-quark-loop amplitudes, it is also necessary to have results for the associated one-loop Feynman integrals, which we also compute in this work. We therefore also introduce the integral family defined through
with inverse propagators defined as:
This adds 18 master integrals to our counting.
Differential Equations and Independent Functions
For the leading-colour light-quark-loop contribution to production, ref. [15] identified the basis of master integrals that brings the corresponding differential equation into -factorised form:
A basis with such a property is known as a “canonical basis”. We call the possible arguments of the d log forms appearing in the system of differential equations in eq. (27) “letters”, and their collection the “alphabet”. We refer the reader to ref. [15] for a complete description of the alphabet, which consists of 152 letters.1
The -factorised form of the differential equation is a crucial feature that facilitates its solution. In particular, let us consider the expansion of the master integrals,
which starts at due to the basis choice of ref. [15]. Notably, the components are rational constants. If we insert this form into eq. (27) we obtain a tower of differential equations at each order in as
Integrating along a path gives
where is a path over phase space that connects the boundary point to the desired evaluation point . Given a boundary value (that can be obtained numerically, e.g. with AMFlow [20, 21]), eq. (30) formally allows one to compute the expansion of the master integrals.
In order to evaluate the master integrals in an efficient way, we will express them in terms of a smaller subset of independent special functions of fixed transcendental weight. We will refer to these functions as pentagon functions and label them as , where the index identifies the transcendental weight of the functions, while enumerates them. To identify them, we proceed weight by weight. At each weight , we first identify a basis of the vector space spanned by the master integrals. We choose the elements of as a linearly independent subset of the -expanded elements of the canonical basis, as is standard (see e.g. refs. [26, 42, 43]). The components of the master integrals are related to such basis by the relation
where is a matrix of rational numbers. By inserting eq. (31) into eq. (29), we get the system of differential equations satisfied by the linearly independent master integrals
where the are rational numbers, which implies the recursive formula, along the same path as in eq. (30):
Not all elements of are new, however. In practice, there exist combinations of the that can be written as polynomials of lower-weight functions and values. We call the elements that are irreducible modulo products of lower-weight functions the pentagon functions , identifying them using the procedure of ref. [25]. As such, all can be written as a polynomial of the and where . We note that, conventionally, we choose the zeta value of weight to be a pentagon function, i.e. it is one of the .
In summary, can be expressed as linear combinations of pentagon functions of the corresponding weight and products of lower-weight pentagon functions. The corresponding relations can be found in the ancillary files attached to this publication.
Pentagon Function Construction
In this section we carry out the construction of section 3 weight by weight, up to weight 4. The numbers of linearly independent and irreducible functions at each weight are given in table 1. At weights 1 and 2, we construct explicit representations for the pentagon functions in terms of logarithms and dilogarithms. At weights 3 and 4, we instead use one-fold integral representations that follow from eq. (33). Many of the steps in our construction are informed by the analytic structure of the differential equation. We thus undertake an analysis of the alphabet on physical phase space, which guides the construction at each weight as well as a study of the analytic structure of the nested square root which has strong implications for our choice of integration path at weights 3 and 4.
| 1 | 7 | 7 |
| 2 | 33 | 17 |
| 3 | 103 | 70 |
| 4 | 141 | 120 |
Table 1. Number of elements of the basis of the space spanned by , eq. (31), and of the pentagon functions , the elements irreducible modulo products of lower-weight functions, at each weight . For , both counts include the zeta value .
Behaviour of the Alphabet on Physical Phase Space
Up to weight 4, the functions constituting our basis can be expressed in terms of the 143 “relevant” letters introduced in ref. [15] (9 further letters contribute only at higher weights). The letters can be categorised depending on their properties under Galois transformations: we define as even letters the ones that have trivial Galois transformations, while we define as odd letters the ones picking up a sign under such transformations. Since these letters will enter the arguments of our special functions, it is particularly important to know their behaviour when evaluated over physical phase-space points, as it determines whether logarithmic branch cuts involving the letters can be kept away from the physical region. In practice, proving sign properties for every letter is an involved task. We therefore sidestep it by studying the letters numerically over physical phase space. The following observations are therefore empirical.
We find that the letters can be divided in two sets:
: letters with fixed sign, either positive or negative. All the even letters fall into this category, and part of the odd letters.
: letters that are a pure phase. All the odd letters that do not have a definite sign fall in this category, and thus lie on the unit circle on the complex plane.
Importantly, we find that none of the purely real letters can vanish over phase space, a feature which is greatly helpful for numerical stability when solving the differential equation. For example, for the sign in can be chosen once and for all such that the logarithm is real-analytic over the whole physical region. To understand the branch-cut structure of special functions having pure-phase letters as an argument, it is important to know which of the elements of can take any possible phase. We find that for only 6 elements of does the image of the physical region cover the whole unit circle. These are given by
For the other elements of , the image of the physical region is a proper arc of the unit circle, which does not contain one of or . This property implies, for example, that one can choose the sign of such that the branch cut is not crossed by any path in the physical region. In contrast, for the letters in no such choice exists: whatever the position of the branch cut, some physical paths cross it. These properties are used throughout the construction. At weight 2, they determine which combinations of dilogarithms are continuous over phase space, while at weight 4, they inform the technique used to construct the anti-derivatives of . A full categorisation of the letters can be found in appendix § 8.§ 1.
Weights 1 and 2: Special Function Representations
We now discuss our construction of the pentagon functions at low weights. For weights 1 and 2 specifically, as is standard, we find representations in terms of well-known special functions: logarithms and dilogarithms. Explicit expressions can be found in ancillary files that we detail in § 5.3.
At weight 1, the span a 7-dimensional space, and all 7 of the are irreducible and hence can be identified as pentagon functions. The analytic expression for the weight 1 pentagon functions can be straightforwardly obtained by integrating eq. (33) and obtaining logarithms of the letters up to a possible transcendental constant. The transcendental constants can be fixed, for example, by comparing the evaluation of the analytic expressions at different phase space points with numerical ones obtained with public tools such as AMFlow [20, 21]. In this way, we define a basis of weight 1 pentagon functions as
We note that the scale enters only and . It arises from expanding the overall factor of an -loop integral, which multiplies lower orders of the expansion. Naturally, a basis of simpler combinations of logarithms is available and the particular ones in eq. (35) arise because we choose our basis to match Feynman integral components. Importantly, the arguments of the logarithms appearing in eq. (35) never cross the branch cut of the logarithm when considered over physical phase space.
At weight 2, the corresponding construction is more intricate. We again begin by counting: the weight 2 master integrals can be written in terms of 17 irreducible weight 2 pentagon functions and products of the weight 1 pentagon functions and can be entirely expressed in terms of polylogarithms. We construct analytic expressions for the weight 2 functions in terms of logarithms and dilogarithms making use of symbol-based techniques [44, 45, 46, 47]. Specifically, we follow the procedure of ref. [46], which has frequently been applied to Feynman integrals (see e.g. refs. [48–51]).
One of the key steps in the procedure of ref. [46] is to identify a set of dilogarithms that can be used to express the weight 2 functions, modulo products of logarithms. To this end, one starts by identifying the possible arguments such that the symbol of the corresponding dilogarithm can be expressed in terms of letters of the alphabet. In order to avoid terms that live outside of the alphabet, the approach of ref. [46] searches for arguments with exponents that satisfy
for some other exponent vector and constants and . Once such a set is identified, an analytic representation of the weight 2 function can be easily constructed using linear algebra at the symbol level, supplemented by numerical evaluations to fix possible constant terms.
In practice, we find that the set of dilogarithms with arguments simultaneously satisfying eq. (36) and of sufficiently low degree to be computationally tractable is insufficient to completely span the space of product-projected weight 2 functions modulo products of logarithms and constants. To remedy this, we introduce “auxiliary letters” built by decomposing the odd letters into numerator and denominator factors. Such auxiliary letters allow the procedure of ref. [46] to succeed at much lower degree. Specifically, we restricted the search to candidates whose numerator and denominator have mass-squared dimension at most 4. As an example of an auxiliary letter, we split the odd letter as
A complete list of all the auxiliary letters and their definition can be found in appendix § 8.§ 2. Importantly, the auxiliary letters do not enlarge the space of d log forms spanned by the alphabet. This follows from the well-known observation that the Galois invariant product of numerator and denominator factors gives products of even letters. For instance, the relation implies that
In the same way, up to the choice of branches, every auxiliary letter factorises over the even and odd letters of the original alphabet.
In our application of the procedure of ref. [46], we put particular care into the selection of a basis of a set of dilogarithms such that we could represent each weight 2 function with a single representation that covers all of physical phase space. This implies selecting arguments for the functions that manifest the analytic properties of the pentagon functions. Specifically, the weight 2 pentagon functions are continuous over phase space, so we looked for ways to build (or simple combinations thereof) that also satisfy this property. Since every auxiliary letter factorises over the letters of the alphabet, every dilogarithm argument, including those containing auxiliary letters, can be written, up to the choice of branches, as
Here is a constant, which also absorbs the constant phases arising from fractional powers of negative letters and from the choice of branches. For the arguments we use, no letter of appears. The fractional powers of the pure-phase letters are therefore continuous over phase space. Moreover, as we are working at weight 2, the collection of complex factors that arise is greatly restricted. Naturally, complex values arise due to the presence of square roots in the letters. In practice, the set of square roots consists of
Out of this set, only is imaginary in the physical region, while the other six roots are real. Importantly, this means that in the following we can identify complex conjugation with the Galois transformation , which helps to organise our analysis.
In practice, we find that we can classify the different arguments into three categories
. In this case is times a positive quantity, so its phase is that of and is fixed over phase space. For the arguments that we use, is real, so is real with a fixed sign. This allows us to introduce dilogarithms in our basis of weight 2 functions that, over phase space, never cross the branch cut starting at . In particular, for , we make use of while, for , we make use of .
and . In this case, is a pure phase, a complex number with modulus 1: over phase space, is thus continuous, since the unit circle meets the branch cut only at , where is continuous (see Figure 2). However, while all the complex letters that arise in are odd under , in general is not. For weight two functions that are odd under the associated Galois transformation, it is natural to choose a basis for the piece which is also odd. To this end, we explicitly make use of the function
Figure 2. Branch cut of (in orange) together with the two kinds of complex arguments of the dilogarithms appearing in the pentagon functions (in blue) pure phases, which stay on the unit circle, and arguments , whose modulus also varies over phase space (dashed, shown for one value of ).
which is manifestly odd, since maps to . On the unit circle, , one has from which we can see that is purely imaginary and continuous on the whole unit circle.
, . In this case, has the form . Unlike the previous two cases, the form of does not prevent it from reaching the branch cut (see Figure 2). Rather than limiting ourselves to arguments where this does not happen, we use
which is continuous for any , since the discontinuity of across is purely imaginary. Moreover, is even under , which maps to and it therefore is a natural building block of Galois even functions.
In this way, each weight 2 pentagon function is given by a single expression in terms of , and , products of logarithms and constants, which is continuous over the whole physical region and so needs no analytic continuation. These expressions are provided in the ancillary files (see § 5.3) and serve as input to the one-fold integral representations at weights 3 and 4.
The Nested Square Root and Paths in Physical Phase Space
Beyond weight 2, our approach to solving the differential equations rests upon representing the solution as a one-fold integration along a path, as in eq. (30). At weights 3 and above, the iterated integration representation in eq. (30) involves letters that depend on the nested square roots . These nested square roots turn out to have a non-trivial analytic structure that influences our choice of integration path. In this subsection, we therefore analyse this structure. We first discuss how the physical phase space constraints fix the signs of the arguments of the nested roots, and then identify where the nested roots fail to be analytic, which restricts the integration paths we may use.
To begin, we observe that the signs of are fixed over physical phase space. From eq. (16), we have
which is non-positive throughout physical phase space as by eq. (7) and the remaining factors of are squares. Thus, for , and have opposite signs.
Moreover, the non-positivity of implies that and we can therefore see that on physical phase space
In particular, is real throughout physical phase space, while both and are purely imaginary.
Let us now determine where the inner root vanishes, which is where it could give rise to an algebraic branch point. Using , its argument can be written as a sum of two manifestly non-negative quantities,
Importantly, if the right-hand-side of eq. (45) is to vanish, each of these non-negative quantities must vanish separately. Given our physical phase space assumptions, the factors and of the second term are strictly positive on physical phase space. (For the latter, note that .) Consequently, within physical phase space we have that
We will denote this surface as . Notably, we see that is a codimension two surface, as it involves two constraints. We therefore do not need to design any special procedure to avoid this branch point surface, since generic paths avoid it.
Despite this comparatively simple behaviour of the inner and outer square roots individually, it turns out that their interplay leads to a non-trivial analytic structure of the nested roots on physical phase space. We sketch the general features in Figure 3. The surface plays a special role in the analytic structure of the nested roots, although this surface is not a Landau singularity and consequently is not a letter in the alphabet. Indeed, on this surface we have that

Figure 3. (a) Two-dimensional slice of phase space. (b) Nested roots along the straight line path between and . Where the path crosses , has a kink and is discontinuous.
Thus, away from , exactly one of and vanishes on . Which of the two vanishes is determined by the sign of . It is therefore useful to distinguish the two regions
where vanish respectively. Figure 3a shows these surfaces on a two-dimensional slice of physical phase space, in which all invariants other than and are held fixed. In this slice, the surface becomes the diagonal. Since has codimension two, it reduces to a single point on the diagonal, which separates (solid) from (dotted).
While the vanishing of the nested roots on is manifest, their behaviour in a neighbourhood of these surfaces is more subtle. We first analyse the behaviour of the two nested roots in a neighbourhood of . Consider . By eq. (43), we can rewrite it in the form
Figure 3: Structure of the non-analyticities of nested square roots on a two-dimensional slice of physical phase space, where all invariants other than and are fixed. The non-analyticities occur along , one half of the surface. This disqualifies integration paths that cross (orange), while allowing those that bypass it (green).
Given the sign properties of the various factors over physical phase space, this representation greatly simplifies taking the principal square root. Indeed, the only quantity that is not of definite sign is and we can write
where we use the principal branch of the square root of . This representation holds wherever , in particular in a neighbourhood of . As is analytic along , the appearance of the absolute value in eq. (50) tells us that is not smooth there. Thus, develops a kink across . Importantly, this behaviour flows into the behaviour of the algebraic root . Indeed, combining its definition eq. (19) with eq. (50), one finds
We therefore see that is discontinuous across .
Importantly, the nested roots are much better behaved along . As a point on stays away from , both the arguments of the inner and outer root when calculating are away from the branch locus. Hence, within a neighbourhood of a point on , is analytic. Thus, by eq. (19), as is non-zero, we see that is also analytic along . To exemplify this behaviour, in Figure 3b, we depict the nested roots over the straight-line path from to (orange) that crosses . We note that along the path in Figure 3a that passes through the point on , the square roots are perfectly analytic.
Let us now consider how this analysis affects our solution of the pure integrals. In practice, paths that cross are safe, while paths that cross must be treated with care. This follows as the pure integrals, defined with the principal branch of , are themselves discontinuous across . To see this, note that of the pure integrals, only and involve the nested roots. They are defined as a combination of scalar Feynman integrals which are analytic on as it is not a Landau singularity. The combination itself, however, involves and , which, by eqs. (50) and (51), causes one of the pure integrals to change sign across . Along a path that crosses , eq. (33) instead yields a continuous function, which therefore no longer reproduces the pure integrals beyond the crossing. While paths that cross could in principle be treated consistently,2 we choose to avoid this class of complications by only considering paths that do not cross . In practice, the straight-line paths that we use in § 5 start at a point and do not cross the surface again. In particular, they never cross .
Weights 3 and 4: One-Fold Integral Representations
At weights 3 and 4, we again choose a basis of pentagon functions modulo products from components of the expansion of the Feynman integrals. As is well known, in contrast to weights 1 and 2, the construction of representations of the special functions in terms of polylogarithms is a demanding task. We therefore follow the standard procedure of representing the functions as one-fold integrals along a path in phase space [22]. To this end, we introduce a path between an initial point and a final point , reached at points and respectively. That is,
The explicit construction of the path and the choice of the boundary point are given in § 5. For the moment, we will only require that is a path within physical phase space that does not cross (see § 4.3). The paths we use start at and do not return to the surface ; this is used below at weight 4. Along the path, we abbreviate
so that, e.g., and . Below, we first give the one-fold integral representation at weight 3, and thereafter discuss the one-fold integral representation at weight 4.
Weight 3: Given a path , the linearly independent special functions at weight 3 can be written as a one-fold integral as
where was defined in eq. (32). Naturally, the weight 3 pentagon functions are simply a subset of the linearly independent special functions and so eq. (54) can be used to evaluate them. Importantly, as is a path entirely within the physical region, we can use the , expressed through the , products of the and constructed in § 4.2, in the integrand of eq. (54). Thus, only the boundary values are required, which can be easily obtained from programs such as AMFlow [20, 21].
The representation in eq. (54) is commonly used to compute pentagon functions via numerical integration (see e.g. refs. [22, 23, 24, 25, 26]). This is the approach that we will follow in § 5. As none of the letters vanishes inside the physical region (§ 4.1), very little dedicated analytic work is required for the implementation of most weight 3 pentagon functions. However, at weight 3 we first encounter letters which depend on the nested square root, which requires dedicated study. In particular, issues arise when numerically evaluating a that involves the nested root. Specifically, the analytic expression of such a derivative may involve poles at one of , or . Given that is not a letter, one expects such poles to be spurious. In practice, however, finding a representation of the in which all spurious poles are manifestly absent is non-trivial. Nevertheless, it is relatively straightforward to construct multiple representations, each of which is manifestly regular on one of or .
As an illustrative example, consider the letter , which can be factorised into
where we define the manifestly odd combination through
Let us focus on the factor . If we compute its naively, we find
This representation is manifestly free of a pole on . On , however, its denominator vanishes as , despite this singularity being spurious. Alternatively, some ad-hoc manipulations allow us to rewrite the form as
In contrast to the previous representation, the form in eq. (58) is manifestly regular on . Nevertheless, it contains a corresponding spurious pole on . Thus, by choosing between the two representations according to the region of interest, the spurious singularities can be avoided.
In this example, we have therefore shown that the form can be made regular throughout , at the cost of using multiple equivalent representations. In practice, we perform an analogous ad-hoc analysis for each letter involving nested roots. Choosing the representation appropriate to the relevant branch locus removes spurious numerical singularities and substantially improves the stability of the numerical integration.
Weight 4: At weight 4, we follow the same strategy outlined for weight 3 and we construct an integral representation for the weight 4 functions. Along the path , eq. (33) at weight 4 is an integral over the weight 3 functions. In the notation of eq. (53), we have:
This two-fold integral representation can be expressed in a one-fold representation by exchanging the order of integration over and in the second term:
where we define the function as
As at weight 3, the weight 4 pentagon functions are elements of , so eq. (60) gives them directly.
The integral in eq. (61) can be written in closed form in terms of logarithms of letters, provided their branches are chosen such that they are continuous along every physical path. There are therefore two subtleties to consider: choosing such branches, and evaluating the letters themselves stably near , which we will now discuss. In most cases, as described in the alphabet study in § 4.1, we can choose the branches in this way along any path over physical phase-space. In particular, we can express as:
with the principal branch of the logarithm. All three expressions are continuous along every physical path, as follows from the behaviour of the letters over phase space. If , has a fixed sign over phase space and the ratio is therefore positive throughout. If , is a pure phase and the discussion becomes more subtle. For , the image of the physical region misses one of the points (§ 4.1). The sign is then chosen for each one of these letters in order to avoid the crossing of the logarithmic branch-cut. For the 6 letters listed in eq. (34), the image covers the whole unit circle (§ 4.1), and eq. (62) would have to be supplemented by winding numbers. Instead, for these letters we evaluate directly as the integral of its d log form, eq. (61), using a Chebyshev approximation [30], which we discuss further in section 5.1.
A second subtlety again concerns the nested square roots. Unlike at weight 3, where only d log forms enter, eq. (62) requires the values of the letters themselves, which must be evaluated stably near . Let us return to our example. Its representation in eq. (55) takes the indeterminate “zero over zero” form on as both numerator and denominator vanish. This can be avoided by using the equivalent representation
where denotes making use of the algebraic square root of eq. (19). Importantly eq. (63) manifestly avoids the zero over zero behaviour on . Evaluating using this second representation therefore gives a substantially more stable result in the vicinity of . The same strategy can be applied to other letters as needed, for which we construct suitable alternative representations on an ad-hoc basis.
More generally, we find numerically that none of the vanish on . Thus, despite the apparent singular behaviour of some naive representations, the letters themselves are regular on . We do not have an algebraic proof of this statement, but have verified it numerically for the letters relevant to our implementation. This property is central to the numerical stability of our approach.
Numerical Evaluation
In section 4 we discussed our analytic construction of a set of pentagon functions to express the planar light-quark-loop integrals for at two loops. In this section we discuss our implementation of their numerical evaluation. For weights one and two we simply implement their analytic forms and evaluate them numerically. Most of our discussion will therefore centre on weights three and four.
Implementation of One-Fold Integrals
The weight-three and weight-four functions are obtained from eqs. (54) and (60). To avoid the analyticity issues discussed in section 4.4 we choose a boundary point on the surface . Concretely, we choose the rational phase-space point
in units of , corresponding to and . At this point we evaluate the canonical master integrals with AMFlow [20, 21] and extract the boundary values to 70 digits.
We choose our integration path to be a straight line starting at and ending at , i.e.
Such a line never returns to the surface, and the nested roots are analytic along it for . As is well-known, straight line paths such as eq. (65) do not always stay inside the physical region as they may violate the Gram constraint . For lines that do violate this constraint, we reach in two straight segments, , through an intermediate physical point generated using Rambo [52]. Importantly, we require to lie on the same side of as , such that the two-segment path is guaranteed not to cross . In practice, only 0.6% of the points in our phase-space sample require a two-segment path.
To evaluate the one-fold integrals, we use the PentagonFunctions++ approach of tanh–sinh quadrature [53], implemented in Boost.Math [54]. Such a quadrature rule exhibits slow convergence when a singularity of the integrand approaches the interior of the interval from the complex plane. As no letter vanishes inside the physical region (§ 4.1), such singularities never lie on the path itself, but they can come close to it. We took particular care in two cases. The first is the inner root of the nested root. Every letter that involves the nested roots depends on and since
their d log forms contain in the denominator. The second is the even letter . Every even letter gives poles of its d log form at its zeros, but only produced zeros close to the integration interval in our study. Along a straight segment, both and are polynomials of degree six in with real coefficients, and we find that neither vanishes on generic paths on physical phase space. As already noted, both are of fixed sign over physical phase space and therefore zeros near the segment come in complex-conjugate pairs . This structure allows us to easily remedy the instabilities they induce. For each segment, we compute the roots of both polynomials with the Durand–Kerner method [55]. For every pair with and , we split the integration interval at . After the split, the nearly real singularity lies close to an endpoint of each subinterval, where the tanh–sinh rule handles it well. We use .
A further issue where we place some care is in the numerical evaluation of the nested roots themselves. A first simple observation is that the evaluation of the nested root using the representation in eq. (19) can be unstable in regions where is small. In practice this can be remedied by using eq. (51), which avoids the problematic division by a small . Beyond this, a stability issue arises in the region , where a naive evaluation of the nested roots can suffer from loss of precision. Specifically, if is small, one has
and, depending on the sign of , the corresponding expressions for can therefore involve a subtraction of nearly equal quantities. For definiteness, consider the case , since the case can be obtained by symmetry. We can then replace the potentially unstable combination by
As is a simple, factorized polynomial expression and the denominator on the right-hand side involves no cancellation, this representation is substantially more stable than the direct evaluation of the expression on the left.
Beyond these dedicated tricks, in order to further improve numerical stability we find it useful to numerically evaluate some of the d log forms in higher precision. Specifically, when the library is built with support for extended-precision arithmetic, using the QD library [56], as for the quadruple-precision evaluations of § 5.2, we evaluate a subset of the d log forms at the quadrature nodes in quadruple or octuple precision, and convert the result back to the working precision before it enters the integrand. All other d log forms, and the rest of the integrand, are evaluated in the requested working precision. Since only a few d log forms are affected, the cost of this promotion is small compared to the cost of promoting the entire pentagon function to higher precision.
Finally, for the six letters of eq. (34) we evaluate directly from its integral representation, eq. (61), as motivated at the end of § 4.4. For this we use the adaptive Chebyshev approximation [30, 31, 57], following ref. [30], which we have implemented in C++ for all three working precisions. On we approximate the integrand by a piecewise Chebyshev interpolant, which can be integrated analytically. This yields a piecewise polynomial approximation of that can be evaluated at any point of the path and so plays the role of a closed-form antiderivative. Each piece starts with 9 Chebyshev nodes, which are doubled up to 129 until the coefficients have decayed below the target accuracy; otherwise the piece is bisected. The interpolants are built once per phase-space point, before the outer integration over , so that at each quadrature node costs only a polynomial evaluation. As the d log forms have no spurious singularities on physical phase space, the construction converges without further intervention.
Rescue System, Performance and Validation
In this section we discuss our precision rescue system, and the validation and performance of our numerical implementation of the pentagon functions.
To validate the implementation and assess its performance, we study our implementation on a physical sample of phase-space points. Our pool of 60000 phase-space points is generated with MATRIX [58] for the subprocess at TeV, distributed according to the LO phase-space weight. Of these, 10000 have been generated for the inclusive cross section with standard PDG values for the top quark and Higgs boson masses, 10000 in the same setup but probing the phase space in the tail of the distributions, and 40000 by varying the masses of the top quark and Higgs boson, one at a time, by GeV. We validate our implementation by taking a random subset of points from this sample and compare the values of all pentagon functions to those obtained by a high-precision numerical evaluation of the master integrals using AMFlow. We find excellent agreement between the two approaches.
Over such a physical phase space, a small set of evaluations lose a significant amount of precision and we therefore implement a precision rescue system that detects such phase space points and automatically repeats the calculation at higher precision. We detect unstable points by making use of two error measures on our results. For definiteness, we discuss the case of weight 4 functions, and decompose the expression in eq. (60) into its three contributions, according to
with
A similar expression holds for weight 3 functions, see eq. (54), where . One source of numerical error originates from the numerical integration to obtain , which is estimated by the integrator. We denote the corresponding uncertainty estimation with and define
Another source of precision loss stems from large cancellations between the three individual contributions in eq. (60). To catch points in which large cancellations between the three contributions are dominant we estimate their uncertainty by defining
where we have chosen . With these two error estimates in hand, we estimate the numerical error on a given pentagon function as
If the estimated error satisfies , the rescue system is triggered, which, for double and quad precision evaluations, will re-evaluate the entire function in higher precision. To address the performance and accuracy of our implementation, we evaluate the entire sample of phase-space points and compare the results in double precision with those obtained in quadruple precision. We define
where the superscripts “quad” and “double” denote evaluations in quadruple and double precision, respectively. Therefore, is a measure for the numerical stability of our implementation. Figure 4 shows the distribution of the number of correct digits, , over this sample. For the bulk of the points we find at least 10 correct digits. Only 1.1% of the points have fewer than 10 correct digits leading us to conclude that our implementation is suitable for phenomenological application. The rescue system is triggered for 0.3% of the points in this sample. Its cost is included in the timings of table 2.
Figure 4. Distribution of the minimal number of correct digits for all pentagon functions as measured by , given in eq. (76). The special functions are evaluated on 60000 phase-space points.
| Median (ms) | Mean (ms) | Max (ms) | |
| double precision | 415 | 483 | 38827 |
| quadruple precision | 35403 | 42554 | 2319508 |
Table 2. Timings for the evaluation of the full set of pentagon functions at a single phase-space point in double and quadruple precision. Timings are measured on an Intel Core i9-14900K with 32 threads.
Finally, to characterise the performance of the implementation, we report timings in table 2. These are measured on an Intel Core i9-14900K with 32 threads over the same sample of phase-space points.
Ancillary Files
In order to ease use of our analytic results and the pentagon functions code, we provide a series of ancillary files alongside this work. They are organised as follows:
funcs.m: a reference evaluation of the pentagon functions fromPentagonFunctions++at the phase-space point given by and
in units of .
letters.m: provides the letters in terms of kinematical invariants including both the auxiliary letters and the four letters that were not in d log form in ref. [15].
pentagonBasis_weighti.m: provides the substitution which writes each master integral at weight in terms of pentagon functions.
roots.m: expresses the square roots appearing in the alphabet in terms of Mandelstam invariants.
subAnalytic_weighti.m: provides the substitution rule which writes each element of the pentagon basis at weight in analytic form, as a function of the letters.
benchmark.m: notebook evaluating all the master integrals in the phase space point , both using the output of thePentagonFunctions++library and the analytic expressions, comparing them.
Summary and Outlook
In this paper we have presented a basis of special functions, together with an efficient numerical implementation, for the planar two-loop master integrals entering the leading-colour amplitudes for with a closed light-quark loop. Starting from the -factorised differential equations of ref. [15], we expressed the master integrals in terms of a minimal set of pentagon functions whose symbols are built from the alphabet of 152 letters. Functions of weights one and two were obtained analytically in terms of logarithms and dilogarithms while functions of weights three and four were represented as one-fold integrals along paths in phase space. The main new feature of this problem is that the canonical differential equations involve nested square roots. We analysed their behaviour on the physical region and showed that, on the hypersurface , which is not a Landau singularity, develops a kink and a discontinuity. To sidestep this difficulty, we place the boundary point of the integration paths on this hypersurface and construct paths that do not return to it. We also found that natural representations of some letters, and of their d log forms, show spurious zero-over-zero behaviour or spurious poles on this hypersurface, even though the letters themselves are regular there. We remove these by finding representations that manifest the lack of spurious poles along our integration path. In addition, six letters are pure phases whose image over the physical region covers the whole unit circle, so that no choice of branch cut makes the logarithm a global antiderivative of their d log forms. Rather than tracking the number of times the integration path would cross the branch cut of a logarithm or partitioning phase space, we end_mid=1 construct these antiderivatives numerically along each integration path with an adaptive Chebyshev approximation [30]. We implemented the resulting evaluation in C++, as an extension of the PentagonFunctions++ library3 [24, 25, 26]. We show that our approach is sufficiently stable over a sample of physical phase space to be used in phenomenological applications. Moreover, we find that a typical evaluation of the full set of functions takes around half a second.
As the difficulties we encountered stem from the algebraic structure of the alphabet rather than from this particular process, we expect that the strategy, namely paths adapted to the non-analytic loci of the roots, equivalent letter representations chosen by region, and Chebyshev approximation for letters whose phase covers the whole unit circle, will carry over to other multi-scale problems with similar structure (see e.g. [36, 37, 38, 40]). Our results provide a building block for precise studies of at NNLO. The same integrals also enter the two-loop amplitudes for , so our results are relevant beyond production. Since the evaluation is fast enough for phase-space integration, our results mark an important stepping stone towards the full computation of the double virtual contribution of the NNLO corrections to .
Acknowledgements
We thank Stefan Kallweit for generating the phase-space sample used in our studies. We thank Vasily Sotnikov for assistance with the PentagonFunctions++ code. We thank Sebastian Pögel for assistance in constructing d log forms. AI tools assisted in the final stages of editing and C++ implementation of this work. MK acknowledges financial support through the DGAPA-PAPIIT grant IA103126 at UNAM, and the SECIHTI “Ciencia Básica y de Frontera” grant CBF-2025-I-615. The work of SD received support from the FWO (contract No. 1227426N) and the European Union (ERC, MultiScaleAmp, Grant Agreement No. 101078449). Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.
The Alphabet
Properties of Letters Over Phase Space
In the following, we list the letters appearing in the alphabet according to their behaviour over phase-space. We report the properties of all letters, including both the relevant letters that we make use of in this work, and irrelevant letters that only arise at .
letters with fixed sign:
We note that in ref. [15] the letters are the square roots , , , , , , and . We use their squares, which are polynomials of fixed sign.
letters that are a pure phase:
The underlined letters are elements of , the ones whose image over the physical region covers the full unit circle in the complex plane, making it impossible to define in such a way that a physical path cannot cross its branch-cut, as described in § 4.1.
Auxiliary Letters
To construct analytic expressions for the weight-2 functions we introduced an additional set of auxiliary letters, which further breaks down the original alphabet. The full set of auxiliary letters that we employ is defined through
where
We note that, similar to eq. (38), logs of auxiliary letters can be expressed in terms of logs of alphabet elements.
References
- [1]CMS collaboration, Observation of tt̄H production, Phys. Rev. Lett. 120 (2018) 231801 [1804.02610].
- [2]ATLAS collaboration, Observation of Higgs boson production in association with a top quark pair at the LHC with the ATLAS detector, Phys. Lett. B 784 (2018) 173 [1806.00425].
- [3]ATLAS collaboration, A detailed map of Higgs boson interactions by the ATLAS experiment ten years after the discovery, Nature 607 (2022) 52 [2207.00092].DOI
- [4]CMS collaboration, A portrait of the Higgs boson by the CMS experiment ten years after the discovery, Nature 607 (2022) 60 [2207.00043].DOI
- [5]M. Cepeda et al., Report from Working Group 2: Higgs Physics at the HL-LHC and HE-LHC, CERN Yellow Rep. Monogr. 7 (2019) 221 [1902.00134].
- [6]ATLAS, CMS collaboration, Highlights of the HL-LHC physics projections by ATLAS and CMS, 2504.00672.
- [7]S. Catani, I. Fabre, M. Grazzini and S. Kallweit, tt̄H production at NNLO: the flavour off-diagonal channels, Eur. Phys. J. C 81 (2021) 491 [2102.03256].
- [8]S. Catani, S. Devoto, M. Grazzini, S. Kallweit, J. Mazzitelli and C. Savoini, Higgs Boson Production in Association with a Top-Antitop Quark Pair in Next-to-Next-to-Leading Order QCD, Phys. Rev. Lett. 130 (2023) 111902 [2210.07846].DOI
- [9]S. Devoto, M. Grazzini, S. Kallweit, J. Mazzitelli and C. Savoini, Precise predictions for tt̄H production at the LHC: inclusive cross section and differential distributions, JHEP 03 (2025) 189 [2411.15340].
- [10]R. Balsach et al., State-of-the-art cross sections for tt̄H: NNLO predictions matched with NNLL resummation and EW corrections, 2503.15043.DOI
- [11]C. Biello, C. Savoini, C. Signorile-Signorile and M. Wiesemann, Next-to-next-to-leading order event generation for \(t\bar{t}H\) production with approximate two-loop amplitude, JHEP 06 (2026) 070 [2603.06143].DOI
- [12]G. Wang, T. Xia, L.L. Yang and X. Ye, Two-loop QCD amplitudes for \(t\bar{t}H\) production from boosted limit, JHEP 07 (2024) 121 [2402.00431].arxiv.org/abs/2402.00431
- [13]B. Agarwal, G. Heinrich, S.P. Jones, M. Kerner, S.Y. Klein, J. Lang et al., Two-loop amplitudes for \(t\bar{t}H\) production: the quark-initiated \(N_f\)-part, JHEP 05 (2024) 013 [2402.03301].
- [14]X. Chen, M. Grazzini and S. Kallweit, Two-loop QCD contribution to the quark-initiated production of a top-antitop quark pair and a Higgs boson at the LHC, 2610.40266.
- [15]F. Febres Cordero, G. Figueiredo, M. Kraus, B. Page and L. Reina, Two-loop master integrals for leading-color \(pp \to t\bar{t}H\) amplitudes with a light-quark loop, JHEP 07 (2024) 084 [2312.08131].DOI
- [16]F. Moriello, Generalised power series expansions for the elliptic planar families of Higgs + jet production at two loops, JHEP 01 (2020) 150 [1907.13234].
- [17]M. Hidding, DiffExp, a Mathematica package for computing Feynman integrals in terms of one-dimensional series expansions, Comput. Phys. Commun. 269 (2021) 108125 [2006.05510].arxiv.org/abs/2006.05510
- [18]X. Liu, Y.-Q. Ma and C.-Y. Wang, A Systematic and Efficient Method to Compute Multi-loop Master Integrals, Phys. Lett. B 779 (2018) 353 [1711.09572].DOI
- [19]X. Liu and Y.-Q. Ma, Determining arbitrary Feynman integrals by vacuum integrals, Phys. Rev. D 99 (2019) 071501 [1801.10523].
- [20]X. Liu and Y.-Q. Ma, AMFlow: A Mathematica package for Feynman integrals computation via auxiliary mass flow, Comput. Phys. Commun. 283 (2023) 108565 [2201.11669].DOI
- [21]R.-J. Huang, X. Liu and Y.-Q. Ma, AMFlow 2.0: significant algorithmic and software improvements for Feynman integral evaluation, 2607.08477.
- [22]S. Caron-Huot and J.M. Henn, Iterative structure of finite loop integrals, JHEP 06 (2014) 114 [1404.2922].
- [23]T. Gehrmann, J.M. Henn and N.A. Lo Presti, Pentagon functions for massless planar scattering amplitudes, JHEP 10 (2018) 103 [1807.09812].DOI
- [24]D. Chicherin and V. Sotnikov, Pentagon Functions for Scattering of Five Massless Particles, JHEP 20 (2020) 167 [2009.07803].
- [25]D. Chicherin, V. Sotnikov and S. Zoia, Pentagon functions for one-mass planar scattering amplitudes, JHEP 01 (2022) 096 [2110.10111].DOI
- [26]S. Abreu, D. Chicherin, H. Ita, B. Page, V. Sotnikov, W. Tschernow et al., All Two-Loop Feynman Integrals for Five-Point One-Mass Scattering, Phys. Rev. Lett. 132 (2024) 141601 [2306.15431].
- [27]A. Huss, J. Huston, S. Jones, M. Pellen and R. Röntsch, Les Houches 2023 – physics at TeV colliders: Report on the Standard Model precision wishlist, 2504.06689.arxiv.org/abs/2504.06689
- [28]T. Armadillo, R. Bonciani, S. Devoto, N. Rana and A. Vicini, Evaluation of Feynman integrals with arbitrary complex masses via series expansions, Comput. Phys. Commun. 282 (2023) 108545 [2205.03345].DOI
- [29]R.N. Lee, A.V. Smirnov and V.A. Smirnov, Solving differential equations for Feynman integrals by expansions near singular points, JHEP 03 (2018) 008 [1709.07525].
- [30]S. Abreu, A. Guerreiro and B. Page, Chebyshev Approximations of Feynman Integrals for Collider Physics, 2607.02411.
- [31]Y. Liu and Y. Zhang, CHESS: CHEbyshev p𝑆eudo-Spectral transport for Feynman integral differential equations, 2606.26691.arxiv.org/abs/2606.26691
- [32]P. Petit Rosàs and W.J. Torres Bobadilla, Fast evaluation of Feynman integrals for Monte Carlo generators, JHEP 09 (2025) 210 [2507.12548].arxiv.org/abs/2507.12548
- [33]S. Badger, M. Becchetti, C. Brancaccio, M. Czakon, H.B. Hartanto, R. Poncelet et al., Double virtual QCD corrections to tt̄+jet production at the LHC, JHEP 05 (2026) 044 [2511.11424].DOI
- [34]M. Czakon and L. Tancredi, Solution of Canonical Differential Equations for Integrals on Arbitrary Geometries, 2606.30354.arxiv.org/abs/2606.30354
- [35]X. Chen and H. Zhang, Feynman Integrals Meet Second-Order Partial Differential Equations, 2607.18406.arxiv.org/abs/2607.18406
- [36]S. Badger, M. Becchetti, N. Giraudo and S. Zoia, Two-loop integrals for tt̄+jet production at hadron colliders in the leading colour approximation, JHEP 07 (2024) 073 [2404.12325].
- [37]L. Götzfried and A. von Manteuffel, Two Feynman integral families for vector boson fusion at NNLO QCD, 2609.37990.arxiv.org/abs/2609.37990
- [38]M. Becchetti, C. Dlapa and S. Zoia, Canonical differential equations for the elliptic two-loop five-point integral family relevant to tt̄+jet production at leading color, Phys. Rev. D 112 (2025) L031501 [2503.03603].DOI
- [39]M. Becchetti, D. Canko, V. Chestnov, T. Peraro, M. Pozzoli and S. Zoia, Two-loop Feynman integrals for leading colour tt̄W production at hadron colliders, JHEP 07 (2025) 001 [2504.13011].DOI
- [40]S. Bera, C. Brancaccio, D. Canko, F. Ripani and S. Zoia, Two-loop integrals for tt̄γ production at hadron colliders in the leading-colour approximation, 2610.42062.arxiv.org/abs/2610.42062
- [41]B. Page and S. Poegel, “Trager: a package for algebraic integration for feynman integrals.”
- [42]S. Badger, H.B. Hartanto and S. Zoia, Two-Loop QCD Corrections to Wbb̄ Production at Hadron Colliders, Phys. Rev. Lett. 127 (2021) 012001 [2102.02516].DOI
- [43]T. Gehrmann, J. Henn, P. Jakubčík, J. Lim, C.C. Mella, N. Syrrakos et al., Graded transcendental functions: an application to four-point amplitudes with one off-shell leg, JHEP 12 (2024) 215 [2410.19088].DOI
- [44]A.B. Goncharov, Galois symmetries of fundamental groupoids and noncommutative geometry, Duke Math. J. 128 (2005) 209 [math/0208144].DOI
- [45]A.B. Goncharov, M. Spradlin, C. Vergu and A. Volovich, Classical Polylogarithms for Amplitudes and Wilson Loops, Phys. Rev. Lett. 105 (2010) 151605 [1006.5703].DOI
- [46]C. Duhr, H. Gangl and J.R. Rhodes, From polygons and symbols to polylogarithmic functions, JHEP 10 (2012) 075 [1110.0458].
- [47]C. Duhr, Hopf algebras, coproducts and symbols: an application to Higgs boson amplitudes, JHEP 08 (2012) 043 [1203.0454].
- [52]R. Kleiss, W.J. Stirling and S.D. Ellis, A New Monte Carlo Treatment of Multiparticle Phase Space at High-energies, Comput. Phys. Commun. 40 (1986) 359.DOI
- [53]H. Takahasi and M. Mori, Double exponential formulas for numerical integration, Publ. Res. Inst. Math. Sci. 9 (1974) 721.DOI
- [54]Boost.Math contributors, “Boost.Math.” https://www.boost.org/libs/math.
- [55]I.O. Kerner, Ein gesamtschrittverfahren zur berechnung der nullstellen von polynomen, Numer. Math. 8 (1966) 290.DOI
- [56]Y. Hida, X.S. Li and D.H. Bailey, Algorithms for quad-double precision floating point arithmetic, in Proceedings 15th IEEE Symposium on Computer Arithmetic. ARITH-15 2001, (Vail, CO, USA), pp. 155–162, 2001, DOI.DOI
- [57]Y. Liu, A. Matijašić, T. Peraro, Y. Xu, Z. Yang and Y. Zhang, Two-loop six-point planar massless feynman integrals to higher orders, JHEP 07 (2026) 223 [2603.16831].
- [58]M. Grazzini, S. Kallweit and M. Wiesemann, Fully differential NNLO computations with MATRIX, Eur. Phys. J. C 78 (2018) 537 [1711.06631].


