Elliptic Polylogarithms and Basic Hypergeometric Functions

Multiple elliptic polylogarithms can be written as (multiple) integrals of products of basic hypergeometric functions. The latter are computable, to arbitrary precision, using a q-difference equation and q-contiguous relations.


Introduction
There is a wide class of Feynman integrals, mainly related to massless theories, which can be expressed in terms of multiple polylogarithms. More challenging are Feynman integrals, which cannot be expressed in terms of multiple polylogarithms. Evaluating this integrals one encounters elliptic generalizations of (multiple) polylogarithms (EP); examples can be found in Refs. [1][2][3] and in Refs. [4,5]. From a more abstract point of view on (multiple) elliptic polylogarithms, in particular on their analytic structure, see Refs. [6,7] and Refs. [8][9][10][11]. An interesting problem is to find a suitable integral representation and the analytic continuation of EPs and an efficient algorithm for their numerical evaluation.

Elliptic polylogarithms
In Refs. [1][2][3] the following functions, of depth two, are defined: x j y k j n k m q jk . (1) For the sake of simplicity we shall assume that arg(q) = 0. The results derived can be extended by using analytic continuation. Following Ref. [7] we prefer to start from ELi n ; m (x , y ; q) = ∞ ∑ j=1 x j j n Li m y q j , where Li m (z) is a generalized polylogarithm [12]. Furthermore, q ∈ C x with |q| < 1 and y ∈ C with 1 / ∈ q R y. Thus, for x < |q| −1 the series converges absolutely. It is immediately seen that ELi 0 ; 0 (x , y ; q) has a pole at x = 1.
Next, generalizations of higher depth of Eq.(2) are also defined in Ref. [3]. They are defined by which are elliptic polylogarithms of depth 2 l. We have introduced the abbreviation [n] l = n 1 , . . . , n l etc. The following relation holds: Furthermore, for l > 1, we can use In Eqs.(8)-(10) we find recurrence relations where the starting point is always ELi 0 ; 0 (x , y ; q), e.g.

Basic hypergeometric functions
It is easy to show that ELi 0 ; 0 (x , y ; q) can be written in terms of a basic hypergeometric function. Indeed, where the basic hypergeometric series is defined as (α 1 ; q) n . . . (α r ; q) n (β 1 ; q) n . . . (β s+1 ; q) n (−1) n q n (n−1)/2 1+s−r z n , whit β s+1 = q and the q -shifted factorial defined by The basic hypergeometric series was first introduced by Heine and was later generalized by Ramanujan. For the case of interest we will use the shorthand notation If |q| < 1 the φ series converges absolutely for |z| < 1. The series also converges absolutely if |q| > 1 and |z| < |c q|/|a b|. With a = q, b = y q, c = y q 2 and z = x q we have that the basic hypergeometric series converges absolutely if • |q| < 1 and |x q| < 1, • |q| > 1 and |x| < 1, which is a special case of |z| < |c q|/|a b|.
In the following we will always assume that |q| < 1. Indeed, using p = 1/q and (a ; q) n = (a −1 ; p) n (−a) n p n (n−1)/2 , one obtains the following relation, Using Eq.(16) we obtain and the inversion formula The next problem is the continuation of the φ series into the complex z plane and the extension to complex q inside the unit disc. We have two alternatives, analytic continuation and recursion relations. First we present an auxiliary relation that will be useful in the continuation of φ

q-contiguous relations
There are several q -contiguous relations for φ and one will be used extensively in the rest of this work, see Eq. (A.10) of Ref. [13] (see also 1.10 of Ref. [14] 2 ); using a shorthand notation, i.e.
one derives

Analytic continuation
A detailed discussion of the analytic continuation of basic hypergeometric series can be found in Ref. [14] (Sects. 4.2-4.10) and in Ref. [16] (Theorem 4.1). From Ref. [14] we can use the following analytic continuation where |arg(−z)| < π, c and a/b are not integer powers of q and a, b, z = 0. Furthermore, It is worth noting that the coefficients in Eq.(21) are reducible to 1 φ 0 functions, e.g.
etc. A more convenient way to compute the q -shifted factorial is given by The condition |arg(−z)| < π must be understood as follows: The Mellin transform of φ (for z ∈ R, z > 0) is a Cauchy Principal Value. With |q| < 1 the two series in Eq.(21) can be used for |z| > z 0 with z 0 = |c/(a b) q|. If |z| < z 0 we can use Eq. (20); repeated applications transform c into |q n c| until a value of n is reached for which z n = |q n+1 c/(a b)| is such that |z| > z n .
In Sect. (4.8) of Ref. [14] it is shown that extension to complex q inside the unit disk is possible, provided that where ln q = −ω 1 − i ω2. As seen in the complex mrz plane the condition is represented by a spiral of equation It remains to study the convergence for |q| = 1. In Ref. [17] a condition is presented so that the radius of convergence is positive; furthermore, the numbers q with positive radius are densely distributed on the unit circle. Obviously, in our case (a = q, b = y q and c = y q 2 ), q = 1 reduces any ELi function to a product of polylogarithms.

Basic hypergeometric equation
In the context of functional equations the basic hypergeometric series provides a solution to a second order qdifference equation, called the basic hypergeometric equation, see Refs. [16,18].
We will now show how φ can be computed to arbitrary precision, using a theorem proved in Ref. [19].
Theorem 3.1 (Chen, Hou, Mu). Let f (z) be a continuous function defined for |z| < r and d ≥ 2 an integer. Suppose that with a n (0) = w n . Suppose that exists a real number M > 0 such that and The f (z) is uniquely determined by f (0) and the functions a n (z). With we have thus, by the theorem and for |q| < 1 and |c/q| < 1 (from Eq.(29)) we can determine uniquely F(z) by F(0) and the q -difference equation (Eq.(31)), i.e. we define High accuracy can be obtained by computing for N high enough. Application to 2 φ 1 q , y q ; y q 2 ; q , x q requires |y q| < 1.
If |q| > 1 we can use the q -difference equation downword. In this case and A which requires |q/c| < 1. In both cases the q -difference equation determines F without limitations in the z complex plane.
To summarize, we can compute φ by using the sum of the series inside the circle of convergence (and their analytic continuation) or by using the q -difference equation. The advantage in using the latter is no limitation on z but |q| < 1 and |c/q| < 1 or |q| > 1 and |q/c| < 1 .
To be precise, the function defined by Eqs.(31)-(35) is a meromorphic continuation of φ with simple poles located at z = q −n , n ∈ Z * . Seen as a function of α = ln(z)/(2 i π) the continuation shows poles at 2 i π α = n ω where ω = − ln q. What to do when we are outside the two regions of applicability? Instead of using analytic continuation we can do the following: assume that |q| < 1 but |c/q| > 1. We can use Eq. (20); repeated applications transform |c/q| into |q n c| until a value of n is reached for which |q n c| < 1. Similar situation when |q| > 1 and |q/c| > 1 where repeated applications transform |q/c| into |q −n c| until a value of n is reached for which |q −n c| < 1.
It is worth noting that the Φ function satisfies which are the q -shift along x and y.

Poles
Using Eq.(12) and Eq.(31) we conclude that ELi 0 ; 0 (x , y ; q) has simple poles located at Poles in the complex y -plane can also be analyzed by using Eq. (12) and Eq. (20). Indeed, from Eq.(20) we obtain Using Φ x , 1 q ; q = 1 we have a simple pole of ELi 0 ; 0 (x , y ; q) at y = 1/q with residue −x/q (the pole at x = 1/q has residue −y/q). Repeated applications of the q -contiguous relation exhibit the poles at y q = q −n , n ∈ Z * .
Residues can be computed according to the following chain where the "regular" part admits a Taylor expansion around x = 1/q, x = 1/q 2 etc.
Using the series we obtain R 1 = (y q − 1) 1 q , showing the following residues for ELi 0 ; 0 (x , y ; q): The isolation of simple poles in ELi 0 ; 0 (x , y ; q) is crucial in order to compute Elliptic polylogarithms of higher depth.

Elliptic polylogarithms of higher depth
In this Section we show how to compute arbitrary elliptic polylogarithms, in particular how to identify their branch points (their multi-valued component).
The procedure is facilitated by the fact that both the basic hypergeometric equation and the q -contiguous relation allows to isolate the (simple) poles of ELi 0 ; 0 (x , y ; q) with a remainder given by a "+" distribution.
Introducing the usual i ε prescription we obtain a general recipe for computing elliptic polylogarithms "on the cuts" (x and or y real and greater than 1/q). In the following we discuss few explicit examples.
From Eq.(8) we obtain For x q ∈ R the integral is defined when From Eq.(31) we derive From Eq.(45) it follows that no pole of the two φ functions in Eq.(46) appears for z ∈ [0, 1]; therefore the two φ in the r.h.s of Eq.(46) can be evaluated according to the strategy outlined in the previous Sections. Using we can write where the "subtraction" term is If q −2 < x < q −3 we can iterate once more obtaining an additional ln(1 − x q 2 ) etc. The function S defined in Eq.(49) is a "+" distribution which has simple poles in the y -plane. The explicit result is as follows: ELi 1 ; 0 (x , y ; q) = ELic 1 1 ; 0 (x , y ; q) + ELir 1 1 ; 0 (x , y ; q) , where the "cut" part is while the "restr" is The second iteration gives ELi 1 ; 0 (x , y ; q) = ELic 2 1 ; 0 (x , y ; q) + ELir 2 1 ; 0 (x , y ; q) , We continue with other examples.
We can use the same derivation as before obtaining a result similar to the one in Eq.(48) where we replace where Li 2 (z) is the dilogarithm; for ELi n ; 0 (x , y ; q) with n > 2 we obtain polylogarithms. The explicit result is as follows: ELi 2 ; 0 (x , y ; q) = ELic 1 2 ; 0 (x , y ; q) + ELir 1 2 ; 0 (x , y ; q) , where the "cut" part is ELic while the "rest" part is When both n and m are different from zero the derivation requires isolating x -poles and y -poles.
This case requires additional work. Consider the integral When y , q ∈ R the integral is defined for y < 1/q, otherwise it is understood that y → y ± o ε with ε → 0 + . By using Eq. (20) we obtain where φ(c) = φ (a , b ; c ; q , u). Next we replace a = q, b = z 2 y q, c = z 2 y q 2 and u = z 1 x q. If 1/q < y < 1/q 2 we replace z 2 = 1/(y q) in the φ function of Eq.(62) and obtain a logarithmic part as well as a "subtraction" part. Note that b = 1 gives a terminating series, i.e. φ q , 1 ; q 3 ; q , z 1 x q = 1 .
The integrand has poles in the complex z -plane located at Poles of the first series are isolated by using the q -difference equation while those in the second series are isolated by using the q -contiguous relation. All poles are simple as long as none of the ratios x i /x j , y i /y j and x i /y j is equal to an integer power of q.