The expansion of the confluent hypergeometric function on the positive real axis

The asymptotic expansion of the Kummer function ${}_1F_1(a; b; z)$ is examined as $z\to+\infty$ on the Stokes line $\arg\,z=0$. The correct form of the subdominant algebraic contribution is obtained for non-integer $a$. Numerical results demonstrating the accuracy of the expansion are given.


Introduction
The confluent hypergeometric function 1 F 1 (a; b; z) (or first Kummer function also denoted by M (a, b, z)) is defined for complex parameters a and b by 1 F 1 (a; b; z) = We exclude this last case from our asymptotic considerations. The function 1 F 1 (a; b; z) is entire in z and is consequently completely described in −π < arg z ≤ π. The behaviour of 1 F 1 (a; b; z) for large z and fixed parameters is exponentially large in (z) > 0 and algebraic in character in (z) < 0. The well-known asymptotic expansion of 1 F 1 (a; b; z) for |z| → ∞ is given by [3, p. 328] Γ(a) Γ(b) 1 F 1 (a; b; z) ∼ E(z) + H(ze ∓πi ), (1.1) where the formal exponential and algebraic asymptotic series E(z) and H(z) are defined by (1.2) In [3, p. 328], the sectors of validity of (1.1) are given by − 1 2 π + ≤ ± arg z ≤ 3 2 π − , where > 0 denotes an arbitrarily small quantity. In the asymptotic theory of functions of hypergeometric type (see, for example, [6, §2], [7, §2.3]) the expansion (1.1) is given in the sector −π < arg z ≤ π, with the upper or lower sign being chosen according as arg z > 0 or arg z < 0, respectively. In the sense of Poincaré there is no inconsistency between these two sets of sectorial validity. The exponential expansion E(z) is dominant as |z| → ∞ in (z) > 0 and becomes oscillatory on the anti-Stokes lines arg z = ± 1 2 π, where it is of comparable magnitude to the algebraic expansion. In (z) < 0, the exponential expansion is subdominant with the behaviour of 1 F 1 (a; b; z) then controlled by H(ze ∓πi ). The negative real axis arg z = ±π is a Stokes line where E(z) is maximally subdominant. When H(ze ∓πi ) is optimally truncated at, or near, its least term, the exponential expansion undergoes a smooth but rapid transition in the neighbourhood of arg z = ±π; see [3, p. 67] and [7,Chapter 6]. It is clear, however, that (1.1) cannot account correctly for the exponentially small expansion on arg z = π, since it predicts the exponentially small behaviour (|z|e πi ) a−b e −|z| as |z| → ∞. When a − b is non-integer with a and b real, this is a complex-valued contribution whereas 1 F 1 (a; b; −|z|) is real.
The same argument applies on the other Stokes line arg z = 0, where E(z) is maximally dominant. The subdominant algebraic expansion undergoes a Stokes phenomenon as the positive real axis is crossed. According to the second set of validity conditions of (1.1), just above this ray the multiplicative factor in front of the algebraic expansion is (ze −πi ) −a , whereas just below this ray the factor is (ze πi ) −a . There will be a smooth transition (at fixed |z|) between these two expressions. The first set of validity conditions mentioned above is more confusing, since it appears that on arg z = 0 one has the choice of either (ze −πi ) −a or (ze πi ) −a for the multiplicative factors. In the case of real parameters, this predicts a complex-valued contribution from the subdominant expansion, when in fact it clearly must be real-valued.
Although such subdominant terms are negligible in the Poincaré sense, their inclusion can significantly improve the numerical accuracy in computations; see, for example, [1, p. 76]. The details of the expansion of 1 F 1 (a; b; z) on the negative real axis have been discussed in [5]. In this note we complete this discussion by considering the correct form of the subdominant algebraic expansion as z → +∞ on the positive real axis. Theorem 1], the expansion of the Kummer function 1 F 1 (a; b; −x) for x → +∞ was established, which took into account the correct asymptotic behaviour of the exponentially small contribution on the negative real axis. If we replace the parameter a by b − a in [5, (2.11)], we have

The expansion for
where the coefficients A j are given by and M is a positive integer. Here, the dominant algebraic expansion on the left-hand side of (2.1), which has been optimally truncated with index m o given by The expansion on the right-hand side represents the exponentially small contribution, where the coefficients B j , which are related to the A j , are specified below. If a = n, where n is a positive integer, the sum on the left-hand side of (2.1) consists of n terms and so cannot be optimally truncated. In this case, the expansion (2.1) holds with the upper limit of the sum on the left-hand side replaced by n − 1 and with no contribution from the sum involving B j (since sin πa = 0); see [5,Theorem 2]. The coefficients B j are defined by The coefficients G k,j appear in the expansion of the so-called terminant function T ν (z), which is defined as a multiple of the incomplete gamma function by T ν (z) := ξ(ν) Γ(1 − ν, z), ξ(ν) = e πiν Γ(ν)/(2πi). When so that ν ∼ x as x → +∞, we have the expansion on the Stokes line arg z = π of this function given by [2, §5] for j = 0, 1, 2, . . . and positive integer M . The coefficients G k,j are computed from the expansion where the parameter γ j is specified by by (2.5). The branch of w(τ ) is chosen such that w ∼ τ − 1 as τ → 1 and so that upon reversion of the w-τ mapping τ = 1 + w + 1 3 w 2 + 1 36 w 3 − 1 270 w 4 + 1 4320 w 5 + · · · , it is found with the help of Mathematica that the first five even-order coefficients G 2k,j ≡ 6 −2kĜ 2k,j are 1 From this it is evident that the coefficients B j not only depend on a and b but also on α in (2.5), which in turn depends on the particular value of the variable x under consideration.
If we apply Kummer's transformation to the hypergeometric function in (2.1) we obtain Theorem 1. When a is non-integer we have the expansion The expansion on the right-hand side of (2.9) represents the correct form of the subdominant algebraic expansion of 1 F 1 (a; b; x) as x → +∞ on the Stokes line arg x = 0. When a = n, where n is a positive integer, the sum on the left-hand side of (2.9) terminates after n terms (and so cannot be optimally truncated) and the sum on the right-hand side involving the coefficients B j vanishes. In this case, the expansion (2.9) reduces to the standard result in (1.1), where there is no ambiguity caused by the leading factor (e ∓πi x) −a appearing in front of the algebraic expansion.

Numerical examples and concluding remarks
In this section we present some numerical examples to demonstrate the accuracy of the expansion in (2.9). As has already been noted, the coefficients B j depend on the parameters a and b and also on α (see the definition of γ j in (2.7)), which appears in the value of the optimal truncation index m o in (2.3). The value of α clearly is a function of the particular value of x being considered. In Table 1 we show values of the coefficients B j for different a, b and α (based on an integer value of x). In the calculations we have evaluated the coefficients G 2k,j for 0 ≤ k ≤ 6.
In Table 2 we show the values of the quantity F(x) defined by the left-hand side of (2.9) In each case the optimal truncation index m o of the exponential expansion in F(x) was determined by inspection and the value of α determined from (2.3). In the case with a = 1 2 the sum involving the coefficients A j makes no contribution to H M (x). It can be seen that the computed values of F(x) agree well with the subdominant algebraic expansion.
We remark that the so-called Stokes multiplier on the positive real axis (given by the quantity in curly braces on the left-hand side of (2.9)) is equal to cos πa to leading order. From (1.1) and (1.2) and the second set of validity conditions this quantity has the values e πia and e −πia just above and below arg z = 0. A commonly adopted, heuristic rule is that the Stokes multiplier on the Stokes line is given by the average of these two values to leading order, namely cos πa. This agrees with the result stated in (2.9); see also [7, p. 248].
It would be of interest to extend the result of Theorem 1 to the more general Wright function defined by where α, β > 0 with κ = 1 + β − α and a, b finite complex constants. From the asymptotic theory of functions of this type (see [6], [7, § §2.2.4, 2.3]) we find that  Upon application of the above heuristic rule, we would expect the Stokes multiplier associated with the algebraic expansion on arg z = 0 to be equal to cos(πa/α) to leading order. In the simpler case α = β, a possible approach to investigate this conjecture is the integral representation for (b) > (a) > 0, which reduces to that of (Γ(a)/Γ(b)) 1 F 1 (a; b; z) when α = 1.