Digital Library of Mathematical Functions
About the Project
NIST
8 Incomplete Gamma and Related FunctionsIncomplete Gamma Functions

§8.12 Uniform Asymptotic Expansions for Large Parameter

Define

where the branch of the square root is continuous and satisfies \eta(\lambda)\sim\lambda-1 as \lambda\to 1. Then

Also, denote

and

where \mathop{F\/}\nolimits\!\left(x\right) is Dawson’s integral; see §7.2(ii). Then as a\to\infty in the sector |\mathop{\mathrm{ph}\/}\nolimits a|\leq\pi-\delta(<\pi),

in each case uniformly with respect to \lambda in the sector |\mathop{\mathrm{ph}\/}\nolimits\lambda|\leq 2\pi-\delta (<2\pi).

With \mu=\lambda-1, the coefficients c_{k}(\eta) are given by

8.12.9
c_{0}(\eta)=\frac{1}{\mu}-\frac{1}{\eta},
c_{1}(\eta)=\frac{1}{\eta^{3}}-\frac{1}{\mu^{3}}-\frac{1}{\mu^{2}}-\frac{1}{12%
\mu},

where g_{k}, k=0,1,2,\dots, are the coefficients that appear in the asymptotic expansion (5.11.3) of \mathop{\Gamma\/}\nolimits\!\left(z\right). The right-hand sides of equations (8.12.9), (8.12.10) have removable singularities at \eta=0, and the Maclaurin series expansion of c_{k}(\eta) is given by

where d_{{0,0}}=-\tfrac{1}{3},

and \alpha_{3},\alpha_{4},\dots are defined by

8.12.13\lambda-1=\eta+\tfrac{1}{3}\eta^{2}+\sum_{{n=3}}^{\infty}\alpha_{n}\eta^{n},|\eta|<2\sqrt{\pi}.

In particular,

8.12.14
\alpha_{3}=\tfrac{1}{36},
\alpha_{4}=-\tfrac{1}{270},
\alpha_{5}=\tfrac{1}{4320},
\alpha_{6}=\tfrac{1}{17010},
\alpha_{7}=-\tfrac{139}{54\;43200},
\alpha_{8}=\tfrac{1}{2\;04120}.

For numerical values of d_{{k,n}} to 30D for k=0(1)9 and n=0(1)N_{k}, where N_{k}=28-4\left\lfloor k/2\right\rfloor, see DiDonato and Morris (1986).

For error bounds for (8.12.7) see Paris (2002a). For the asymptotic behavior of c_{k}(\eta) as k\to\infty see Dunster et al. (1998) and Olde Daalhuis (1998c). The last reference also includes an exponentially-improved version (§2.11(iii)) of the expansions (8.12.4) and (8.12.7) for \mathop{Q\/}\nolimits\!\left(a,z\right).

A different type of uniform expansion with coefficients that do not possess a removable singularity at z=a is given by

for z\to\infty in \left|\mathop{\mathrm{ph}\/}\nolimits z\right|<\frac{1}{2}\pi, with \realpart{(z-a)}\leq 0 for \mathop{P\/}\nolimits\!\left(a,z\right) and \realpart{(z-a)}\geq 0 for \mathop{Q\/}\nolimits\!\left(a,z\right). Here

and

8.12.20
A_{0}(\chi)=1,
A_{1}(\chi)=\tfrac{1}{2}\chi+\tfrac{1}{6}\chi^{3},
B_{1}(\chi)=\tfrac{1}{3}+\tfrac{1}{6}\chi^{2}.

Higher coefficients A_{k}(\chi), B_{k}(\chi), up to k=8, are given in Paris (2002b).

Lastly, a uniform approximation for \mathop{\Gamma\/}\nolimits\!\left(a,ax\right) for large a, with error bounds, can be found in Dunster (1996a).

For other uniform asymptotic approximations of the incomplete gamma functions in terms of the function \mathop{\mathrm{erfc}\/}\nolimits see Paris (2002b) and Dunster (1996a).

Inverse Function

For asymptotic expansions, as a\to\infty, of the inverse function x=x(a,q) that satisfies the equation

see Temme (1992a). These expansions involve the inverse error function \mathop{\mathrm{inverfc}\/}\nolimits\!\left(x\right)7.17), and are uniform with respect to q\in[0,1]. As a special case,

8.12.22x(a,\tfrac{1}{2})\sim a-\tfrac{1}{3}+\tfrac{8}{405}a^{{-1}}+\tfrac{184}{25515}%
a^{{-2}}+\tfrac{2248}{34\;44525}a^{{-3}}+\cdots,a\to\infty.