(a) I
6
, I
11
(b) I
9
(c) I
36
Figure 3. Factorized integrals.
was used in the integral reduction. In ref. [23], it was proposed to choose the basis such
that the integrals have simple properties. This considerably simplifies their computation,
and allows to obtain the result in a form that is as compact as possible. As we will see in
the next section, in the expansion, all integrals evaluate to multiple polylogarithms. In
general, such Feynman integrals will involve linear combinations of multiple polylogarithms
of varying transcendental weight, and with various prefactors that depend rationally or
algebraically on the kinematics. It is desirable to disentangle the latter, so that such
factors are moved into overall normalizations of the integrals, and such that only functions
of homogeneous weight appear.
Understanding for predicting which integrals have this property came initially from
studies in N = 4 super Yang-Mills. Conjecturally, integrals whose integrands can be
written as a ‘d-log’-form, and hence have constant leading singularities [37], have this
property. The initial examples satisfying this conjecture were massless, planar, finite,
dual conformal integrals. It has since been generalized to more generic integrals within
dimensional regularization [23].
What is important to emphasize is that the basis choice can be done a priori, by
analyzing the loop integrand. In principle, one could classify all integrands having the
desired properties, and then select a linearly independent (under integral reduction) subset.
This can be done algorithmically, see e.g. [38], and is described in detail for two-loop five-
point functions in section 3 of [39].
In practice, it may not be necessary to classify all such integrals, but just to construct
a sufficient number of them. It is possible to construct many ‘d-log’ integrals directly, for
example by iteratively using lower-loop building blocks. See refs. [40, 41] for examples.
Our choice of 61 basis integrals is given in the ancillary file pentabox basis2.txt.
4 Differential equations for the master integrals
To compute the master integrals, the explicit and complicated integration over the loop
momenta can often be avoided for multi-scale integrals by using differential equations in
kinematical invariants, as first demonstrated for the two-loop four-point functions in [22].
We use the integral basis I
j
, with j = 1 . . . 61, discussed in the previous section, and
compute the differential in all variables v
k
, k = 1, . . . 5. We find the following canonical
– 7 –