Bessel integrals in epsilon expansion: Squared spherical Bessel functions averaged with Gaussian power-law distributions
- Department of Physics, Hiroshima University, 1-3-1 Kagami-yama, Higashi-Hiroshima 739-8526, Japan
- Available online 15 October 2013
The high-index evaluation of integrals containing squared spherical Bessel functions is studied.
The integrals arise as spectral averages in multipole expansions of spherical Gaussian random fields.
A high-precision integration technique based on finite Hankel series in epsilon regularization is developed.
An Airy approximation of the integrals is derived using uniform Nicholson asymptotics.
The finite series evaluation is compared with the Airy approximation over an extended range of Bessel indices.
Bessel integrals of type are studied, where the squared spherical Bessel function is averaged with a modulated Gaussian power-law density. These integrals define the multipole moments of Gaussian random fields on the unit sphere, arising in multipole fits of temperature and polarization power spectra of the cosmic microwave background. The averages can be calculated in closed form as finite Hankel series, which allow high-precision evaluation. In the case of integer power-law exponents μ, singularities emerge in the series coefficients, which requires ε expansion. The pole extraction and regularization of singular Hankel series is performed, for integer Gaussian power-law densities as well as for the special case of Kummer averages (a=0 in the exponential of the integrand). The singular ε residuals are used to derive combinatorial identities (sum rules) for the rational Hankel coefficients, which serve as consistency checks in precision calculations of the integrals. Numerical examples are given, and the Hankel evaluation of Gaussian and Kummer averages is compared with their high-index Airy approximation over a wide range of integer Bessel indices l.
- Squared spherical Bessel functions;
- Regularization of Hankel series;
- Gaussian power-law densities;
- Kummer distributions;
- Airy approximation of Bessel integrals
We investigate the Bessel integrals arising in the multipole expansion of isotropic Gaussian random fields on the unit sphere  and . is a squared spherical Bessel (ssB) function of integer index l⩾0,p is a positive scale parameter, a>0,b and ω are real constants, and μ is an integer power-law exponent.
In Section 2, we derive a finite Hankel series representation of squared spherical Bessel functions, which is used to obtain closed analytic expressions for the above integrals suitable for high-precision calculations. The finite Hankel expansion admits term-by-term integration in terms of confluent hypergeometric functions. At integer power-law exponents μ, the confluent functions become singular, so that the Hankel series requires epsilon expansion, see Section 3, where we perform the pole separation of the confluent functions appearing in the coefficients of the Hankel series. In Section 4, we obtain explicit formulas for the regularized Hankel series in closed form, arriving at a numerically efficient finite series representation of the Bessel integrals in terms of confluent hypergeometric functions.
In Section 5, we discuss a special case, where the averaging of the squared spherical Bessel function is carried out with an exponentially cut and modulated power-law density (a=0 in the above integrand). In this case, the confluent functions in the Hankel coefficients are still singular at integer power-law exponents μ but become elementary, so that the regularized Hankel series are elementary functions as well. The Conclusion, Section 6, gives an overview of the results. In Appendix A, we use the ε pole extraction to derive combinatorial identities (sum rules) for the Hankel coefficients, by making use of the fact that the singular ε residuals of the series coefficients cancel one another if the integrals converge.
2. Integrals of squared spherical Bessel functions expressed as finite Hankel series
We study the Bessel average , where the distribution g(k)=kμe-ak2-(b+iω)k is a modulated Gaussian power law with real exponents μ,a>0,b and ω (or a=0 and b>0). The scale parameter p in the argument of the squared spherical Bessel (ssB) function is positive. Spherical Bessel functions are rescaled ordinary Bessel functions of half-integer order, , the index l being a non-negative integer  and . These integrals converge for μ+2+2l>-1, as the ascending series of starts with the power x2l.
We will need an integral representation of a parabolic cylinder function ,
We split the exponential exp(2ipk) in integral (2.1) into real and imaginary parts, definingDexp and Dcos,sin,0 will be considered as analytic continuations in μ of the respective integrals, which converge for Re μ>-1. This continuation is effected by the Kummer function in (2.4) or by (2.5). In the following sections, we study negative integer exponents μ, where the gamma function in (2.3) becomes singular, which can be dealt with by epsilon expansion. In this section, we assume a non-integer power-law exponent μ.
We will employ a finite Hankel representation of the squared spherical Bessel function,and are polynomials in 1/x, explicitly stated in (2.10), (2.11), (2.12), (2.13), (2.14) and (2.15). In the following, we use the Hankel symbol in (2.9) reads0⩽k⩽l,l⩾0, and [i,j] denotes Hankel’s symbol (2.10).
As for the polynomials in (2.9), we define and0⩽k⩽l-1 and l⩾1. Finally, the polynomials in (2.9) readl⩾0.
We substitute the Hankel representation (2.9) of the squared Bessel function into integraland . In this way, we obtain integral (2.16) as linear combination of three finite Hankel series:a2k(l),b2k+1(l), and c2k(l) are stated in (2.12) and (2.14), and (2.15), and the functions Dcos,sin,0 are defined in (2.6), (2.7) and (2.8), with Dexp in (2.5) substituted. We also note that ; a sum is regarded as void if the lower summation boundary exceeds the upper one. We thus obtain the Bessel integral DssB(l,p;μ,a,b,ω) in (2.16) as a finite linear combination of confluent hypergeometric functions, cf. (2.5).
We consider the modulated and exponentially cut power law kμe-(b+iω)k, with b>0 and real ω. The power-law index μ is real and non-integer. (Integer indices will be studied in Section 5.) This is the limit case a=0 in (2.16) and (2.19). We use the shortcut β=b+iω,Reβ>0, so that the Bessel integral (2.16) readsDexp in (2.1) is now elementary,(2.6), (2.7) and (2.8) of this integral are elementary as well,β=b+iω can be complex, with b>0, and the power-law exponent μ in integral (2.21) is usually real, although the functions Dcos,sin,0 will be regarded as analytic μ continuations of the respective integrals. The Hankel series in (2.18) are compiled with these elementary functions, and integral DssB(l,p;μ,0,b,ω) in (2.21) is assembled as stated in (2.17). If the power-law exponent μ is integer, the coefficients Dcos,sin,0 of the Hankel series (2.18) become singular, which requires ε expansion, performed in Section 5 for the Kummer averages (2.21).
In series (2.29), we consider the limit a=0 and b>0, so that D0(μ,0,b,ω)=Γ(μ+1)/(b+iω)μ+1, cf. (2.19). Series (2.29) is thus majorized by a polylogarithm , with z=4p2/(b+iω)2, which converges in the unit disk |z|<1 for arbitrary μ. This can be seen from the asymptotic series coefficients (2.27) and the limit (2.28) of the gamma function. Rapid convergence apparently requires a large parameter b (Laplace asymptotics) or ω (Fourier asymptotics).
In the opposite limit a→∞, we can use (2.20) as estimate for D0(μ,a,b,ω) in series (2.29). In this case, series (2.29) is convergent due to a factor e-k(logk-1) emerging in the series coefficients for large k, cf. (2.27) and (2.28). However, this factor only emerges for k≫l (since the asymptotic limit in (2.27) only applies in this regime), which requires to sum a large number of terms in series (2.29) if the Bessel index l is large. Rapid convergence requires a large exponent a or a large |b+iω| in integral (2.16). This expansion also works if both a and |b+iω| are large and y is moderate, cf. (2.1) and (2.2). In brief, if the exponents a and/or |b+iω| in the integrand (2.16) are large numbers and the index l is low or moderate, the Laplace/Fourier series expansion (2.29) is efficient. The parameter p in series (2.18) and (2.29) can be put equal to one, as it can be scaled into the exponents a,b and ω of the integrand in (2.16).
3. Bessel integrals with integer power-law exponent: epsilon expansion and Hermite residuals
In Sections 2.1 and 2.2, we considered non-integer power-law exponents μ in integral , cf. (2.16), where is a squared spherical Bessel function of integer index l⩾0, and p a positive scale parameter. Here and in Sections 4 and 5, we study integer power-law exponents μ=m, so that is averaged with the distribution kme-ak2-(b+iω)k, with a>0, real b and ω. The k2 factor in the integrand typically stems from a 3D volume integration in polar coordinates . The exponent m can be negative, provided that m+2⩾-2l, so that the integral is convergent. We will calculate integral (2.16) by performing ε expansion for arbitrary integer power-law exponents μ=m⩾-2l-2.
The analytic μ continuations Dcos,sin(p;μ-n,a,b,ω) and D0(μ-n,a,b,ω) of the integrals in (2.6), (2.7) and (2.8), which are required in the Hankel series (2.18), become singular due to poles in the gamma function in (2.3). We put μ=m+ε, where m is an arbitrary integer, and n=m+1+j, so that 1+μ-n=-j+ε. Poles in Dexp(p;μ-n,a,b,ω) (defined by (2.3) or (2.5)) and in its trigonometric components Dcos,sin,0 in (2.6), (2.7) and (2.8) emerge only at integer μ-n=-j-1 with j⩾0.
We start with the ε expansion of integral Dexp(p;μ-n,a,b,ω) in (2.3), and put μ=m+ε and n=m+1+j, where the pole index j is a non-negative integer; no singularities arise for negative integer j. Thus, cf. (2.3),μ-n=-j-1+ε, and we use the shortcut , cf. (2.2). In (3.1), we substitute the ε expansion of the gamma function, assuming integer j⩾0,ε pole,(3.2) is the logarithmic derivative of the gamma function. The constant term in the ε Laurent expansion of Dexp is. The residue in (3.3) reads(3.7) and (3.8),, we consider even integer pole indices j=2n,n⩾0, and note  in (3.5) is thus real for real y.
For odd integer pole index, we put j=2n+1,n⩾0, in the residue (3.5) to findy) and the Hermite polynomial are imaginary. This defines the singular term in (3.5) for odd pole index, which is imaginary at b=0, that is for real y, cf. (2.2).
At very small but finite ε, the regular part in (3.3) can be calculated with high precision asexplicitly given in (3.5), (3.7) and (3.8). This can be more efficient than the numerical differentiation of the Kummer function with respect to the ε parameter in (3.4).
4. Regularized Hankel series
4.1. Pole subtraction
In Section 3, we performed the ε expansion of integral Dexp in (2.1), and calculated the pole term , cf. (3.3). The pole separation and ε expansion of the trigonometric integrals Dcos,sin,0 defined in (2.6), (2.7) and (2.8) can be assembled from the coefficients and in (3.3).
Returning to the Hankel series and in (2.18), we put μ=m+ε. The gamma function in the analytic ε continuations Dcos(p;μ-2k,a,b,ω) and D0(μ-2k,a,b,ω) (defined by (2.3) and (2.6) and (2.8)) stays regular for ε→0 if the non-negative integer k satisfies 2k⩽m. It is singular for 2k⩾m+1. Accordingly, we split the series defining and in (2.18) asDcos(p;m-2k,a,b,ω) and D0(m-2k,a,b,ω) in (2.6) and (2.8) in the first summation like in (2.18). In the second summation in (4.4), we use, instead of Dcos and D0, the regular coefficients and in (4.3), with j=2k-1-m (so that μ-2k=-j-1+ε, cf. (4.1) and (4.2). The brackets in the upper and lower summation boundaries in (4.4) denote the integer part (that is, the largest integer less than or equal to the indicated argument). We also use the convention that a sum is zero if the lower summation boundary exceeds the upper one.
As for series in (2.18), with μ=m+ε and Dsin in (2.7), we note that the gamma function in Dsin(p;μ-2k-1,a,b,ω) stays regular for ε→0 if k satisfies 2k+1⩽m, cf. (2.3) and (2.7), whereas poles emerge if 2k+1⩾m+1. We split the series defining in (2.18) asDsin(p;m-2k-1,a,b,ω) in (2.7). In the second summation, we replace Dsin by in (4.3) and put j=2k-m (since μ-2k-1=-j-1+ε). The pole terms in (4.1) and (4.2) can be ignored in the calculation of integral (4.6) below, as they cancel one another, cf. Appendix A.
l DssB(l) Hankel series DssB(l) Airy approx. 0 7.991354106316505 ×104 7.992338873 ×104 1 7.999250409735869 ×104 8.000256164 ×104 5 8.113326663187338 ×104 8.114670865 ×104 10 8.481208178016143 ×104 8.483594389 ×104 20 1.042557080617444 ×105 1.042268591 ×105 30 1.238911735222280 ×105 1.236931157 ×105 50 1.511142597127369 ×104 1.509695163 ×104 70 1.503953736747144 ×101 1.180405467 ×101 100 8.233597800611870 ×10-8 8.635323282 ×10-9 150 9.739275886661490 ×10-30 4.028033041×10-36
- Full-size table
l DssB(l) Hankel series DssB(l) Airy approx. 0 1.761712987728264 ×1014 1.761713864 ×1014 1 1.761720004433907 ×1014 1.761720881 ×1014 5 1.761818247890288 ×1014 1.761819125 ×1014 10 1.762099042033407 ×1014 1.762099922 ×1014 50 1.770734121286065 ×1014 1.770735073 ×1014 100 1.798359213865244 ×1014 1.798360424 ×1014 200 1.924862478039058 ×1014 1.924865675 ×1014 300 2.233775172960886 ×1014 2.233790888 ×1014 400 3.075282836068884 ×1014 3.075217334 ×1014 600 – 9.669919328 ×1013 800 – 2.505261104 ×1010 1000 – 1.580371521 ×103 1200 – 2.090425596 ×10-8
- Full-size table
4.2. Consistency checks for Hankel series: comparison with Weber integrals and Airy approximationa>0. A finite representation of the spherical Bessel function jl(x) valid for complex argument isDssB(l,p;m+ε,a,b,ω) in (4.6) to its Airy approximation. To this end, we consider an arbitrary kernel function g(k) and writeg(k)=km+εe-ak2-(b+iω)k to recover integral DssB(l,p,m+ε,a,b,ω). We rescale the integration variable in (4.14) with (l+1/2)/p,,(4.15), which admits the integral representation ,  and  for representations of Airy functions by trigonometric kernels. The indicated approximation in (4.18) holds for real argument and large z, so that the cubic term in the argument of the cosine can be dropped (Riemann–Lebesgue lemma). θ(z) is the Heaviside step function. Thus we put in the interval 0<x<1, where ξ(x) is positive, which means to replace the lower integration boundary in integral (4.15) by 1. In the range x⩾1, we introduce a new integration variable, , and substitute (4.18) into the squared Nicholson approximation (4.16), with , to obtainy=x2-1 to integral (4.15), we obtain the Airy approximation of integral (4.14),(4.14), we find the Airy approximation of integral (4.6),1+y=(1+t)2,y=t(2+t), to remove the root in the exponential,(4.21) or (4.22) is numerically tame and turns out to be efficient even at very low Bessel index l, but its accuracy is limited as compared with the Hankel series evaluation in (4.6), cf. Table 1, Table 2 and Table 3. The Airy approximation of integrals containing products of Bessel derivatives is discussed in .
l Re DssB(l) Hankel series Re DssB(l) Airy approx. ImDssB(l) Hankel series ImDssB(l) Airy approx. 0 2.459394994691453 2.456975786 −2.29953729177942 ×101 −2.29993855 ×101 1 2.439023740399505 2.436746712 −2.30369798295024 ×101 −2.30396971 ×101 5 2.176623546402476 2.174578376 −2.34117813788856 ×101 −2.34127427 ×101 10 1.469537577207173 1.467707567 −2.41348458087903 ×101 −2.41348612 ×101 50 −1.49154918284621 ×101 −1.49147674 ×101 −2.66093614013266 ×101 −2.66074707 ×101 100 −3.54240464462342 ×101 −3.54214934 ×101 −3.37196089026216 −3.37281990 300 2.126889250211135 ×101 2.127187079 ×101 −3.07546677806975 ×101 −3.07505100 ×101 500 1.536265267664984 ×101 1.535679165 ×101 2.612978498620274 ×101 2.613253539 ×101 800 1.214846414087418 ×101 1.214373489 ×101 1.484542229293373 ×101 1.484978720 ×101 1000 −1.30411841949000 ×101 −1.30432858 ×101 3.598946584521200 3.593693228 5000 – 4.215571594 ×10-4 – −3.02117450 ×10-3 104 – −2.26686563 ×10-8 – −3.73505755 ×10-8 2×104 – −5.96529699 ×10-18 – 2.147340549 ×10-18
- Full-size table
The Kummer averages (2.21) are a limit case of integral (2.16), where the quadratic term in the exponent of the integrand vanishes, a=0. We can proceed as in Section 4.1, starting with the ε expansion of integral Dcos(p;μ-n,a=0,b,ω) in (2.23), where μ=m+ε and n=m+1+j, with integer j⩾0. As in Section 2.2, we use the shortcut β=b+iω, and write, cf. (2.23) and (4.1),(3.4), the regular constant term of the ε expansion in (5.1) is elementary,ε expansion (3.2) of the gamma function into (5.1). The constant coefficient (5.3) differs from in (5.1) by terms of O(ε).
Analogously, integral Dsin(p;μ-n,0,b,ω) in (2.24) admits the ε expansionε expansion of integral D0(μ-n,0,b,ω) in (2.25) reads, cf. (4.2),β=b+iω.
As mentioned, in contrast to the Gaussian power-law averages (4.6), the regular coefficients of Kummer averages are elementary functions at ε=0. Therefore, it is numerically efficient to perform the regularized Hankel series expansion for integer power-law index m directly at ε=0. The analogue to the series expansion (4.6) is thusin (5.10) are assembled by substituting the regular constant coefficients calculated in (5.3), (5.6) and (5.9) into series (4.7), (4.8) and (4.9) (where we put ε=0). Also required in these series are the non-singular coefficients Dcos,sin,0 listed in (2.23), (2.24) and (2.25). Integral (5.10) converges if m⩾-2l-2, with integer Bessel index l⩾0 and exponent b>0, since jl(x)∝xl(1+O(x2)) and jl(x)=O(1/x) for large argument.
6. Results, discussion & conclusionkme-ak2-(b+iω)k. The constant p in the argument of the Bessel function is a positive scale parameter. The constants a,b and ω defining the modulated exponential are real, and a is positive, ensuring convergence at the upper integration boundary. (The limit case a=0 with b>0 is also convergent, defining Kummer averages worked out in Section 5. Here, we summarize the general case, a>0.) The power-law index m is integer and can be negative, provided that m+2⩾-2l, which is the condition for integral (6.1) to converge at the lower integration boundary. In multipole expansions of spherical random fields used to analyze temperature fluctuations of the cosmic background radiation, the exponents m,a,b,ω and the scale parameter p are kept constant, and the multipole index l varies over a wide range; presently multipoles up to l∼104 are resolvable  and . The complex exponent b+iω generates modulations in the multipole spectrum.
At negative integer z=-j-1,j⩾0, pole singularities emerge in Dexp and thus in the coefficient functions Dcos,sin,0(z). The regularized coefficients in (6.3) read, after pole subtraction,, where(2.4). The argument y is the same as in (6.5). The regularized coefficients are defined for integer j⩾0. The pole subtraction is performed by means of epsilon expansion, cf. Section 3, stated in (3.5). The ε residues of the coefficients Dcos,sin,0 are defined analogous to (6.8) and calculated as in (6.6) with replaced by .
The residues do not enter in the actual calculation of integral (6.1). If we replace the regular coefficients in the Hankel series (6.3) by the residues , then the second summation in each of the three series in (6.3) must vanish identically. (The poles cancel one another because integral (6.1) is finite.) This pole cancellation happens due to combinatorial identities, sum rules satisfied by the series coefficients a2k(l),b2k+1(l) and c2k(l) in (6.3), which are derived in Appendix A. These identities, cf. (A.10) and (A.11), provide useful consistency checks for the series coefficients in high-precision calculations, especially for large Bessel index l; they only involve rational numbers, and are independent of the parameters defining integral (6.1). Finally, the Bessel index l and the power-law exponent m in the Hankel series (6.3) satisfy l⩾0 and the convergence condition m⩾-2l-2 of integral (6.1). It is evident from the summation boundaries in (6.3) (second summation signs), that regularization by pole subtraction (6.8) is required if the Bessel index exceeds m/2. Since the power-law exponent m of integral (6.1) is usually a small integer, this already happens at very low Bessel index in multipole expansions.
The Hankel method summarized in (6.1), (6.2), (6.3), (6.4), (6.5), (6.6), (6.7) and (6.8) is based on finite series, and can be used for integer Bessel index l, allowing safe evaluation in any desired precision. We have also discussed asymptotic techniques and special cases of integral (6.1) providing cross-checks by numerical comparison. In Section 2.3, we studied the Laplace and Fourier asymptotics of integral (6.1), namely the limit a→∞ or |b+iω|→∞, keeping the Bessel index l fixed. This limit, however, is not applicable in multipole expansions, in which the exponents are moderate and the Bessel index becomes large. In Section 4.2, we pointed out a Weber integral defined by exponents m=0 and b=ω=0 in (6.1), which can be expressed by a spherical Bessel function with imaginary argument. Weber integrals with even integer power-law exponent m are obtained as multiple a derivatives or antiderivatives of identity (4.10), cf. . In Section 4.2, we also derived the Airy approximation of integral (6.1), making use of Nicholson’s steepest-descent asymptotics of the squared Bessel function in the integrand of (6.1). In Table 1 and Table 2, we compare Hankel expansion and Airy approximation by listing integral (6.1) for various Bessel indices.
In Sections 2.2 and 5, we studied the special case of Kummer averages, that is integral (6.1) with a=0 and b>0. In this limit case, the confluent hypergeometric functions in (6.5) and (6.7) become elementary power laws; the Hankel series of these averages are explicitly compiled in Section 5. In this section, we also mentioned a Beltrami integral defined by exponents m=-1,a=0 in (6.1), which can be expressed as an elementary Legendre function of the second kind and integer degree l, cf. (5.14). Table 3 compares the Hankel evaluation of Kummer averages with their Airy approximation (4.21), illustrating the accuracy of the Airy approximation over an extended range of Bessel indices.
Appendix A. Combinatorial identities for Hankel coefficients
We consider identity (A.2) in the limit a→0 , b=0,ω→0. The left-hand side of this identity is a polynomial in a as well as ω, since is a polynomial, cf. (3.5) and (3.6). Thus this identity stays valid for a=ω=0, even if integral (4.6) does not converge at the upper integration boundary. (In this limit, integral (2.16) degenerates into a Schafheitlin integral , convergent in the strip -2l-3<Reμ<-1.) In (3.6), we perform the limit |y|→∞, replacing the Hermite polynomial by its asymptotic limit Hj(z→∞)∼2jzj and the Kummer function bya→0 in , cf. (2.2). On substituting (A.4) into (3.5), we obtain are defined as in (4.3). In this way, we find series in (A.3) at a=b=0 asω→0. It is easy to see that vanishes if m is even or if m⩽-3, since all powers of ω in the series are positive. A non-zero limit is obtained only for odd m⩾-1, provided that l⩾(m+1)/2 (so that the sum is not void). In this case, the first term of the series is independent of ω. Thus, for m=-1,1,3,… and l⩾(m+1)/2, we obtain in (4.3), we substitute in (A.5) and perform the limit ω→0 to find and vanish. Accordingly, the Hankel series and in (A.3) vanish for even m, like series discussed above. If m is odd, we find, by substituting (A.8) into (A.3), holds for odd m⩽-3, since the Hankel series vanishes in this case. Thus we find, for m=-3,-5,-7… and l⩾-(m+1)/2 [the latter being the convergence condition m⩾-2l-2 of integral (4.6) for odd power-law exponent m, required for identity (A.2) to hold], the sum rules still vanishes if l<(m+1)/2, but the summations in (A.9) are void in this case, and no new identities are obtained.
It remains to consider the case of odd m⩾-1, with l⩾(m+1)/2, where does not vanish and is given by (A.7). In this case, identity holds according to (A.2), and we find, by substitution of the Hankel series (A.7) and (A.9),m⩾-1 and integer l⩾(m+1)/2, whereas identity (A.10) applies for odd m⩽-3 and l⩾-(m+1)/2. The rational identities (A.10) and (A.11) provide significant numerical consistency checks for the combinatorial coefficients a2k(l),b2k+1(l) and c2k(l) stated in (2.12), (2.14) and (2.15).
- Planck Collaboration et al., Planck 2013 results. I. Overview of products and scientific results, 2013. arXiv:1303.5062.
Multipole fine structure of the cosmic microwave background: reconstruction of the temperature power spectrum
Mon. Not. R. Astron. Soc., 427 (2012), pp. 1363–1383
Scattering Theory of Waves and Particles
Springer, New York (1982)
Formulas and Theorems for the Special Functions of Mathematical Physics
Springer, New York (1966)
A Treatise on the Theory of Bessel Functions
Cambridge University Press, Cambridge (1996)
Polylogarithms and Associated Functions
North-Holland, Amsterdam (1981)
Integral representations for products of Airy functions
Z. Angew. Math. Phys., 46 (1995), pp. 159–170
Electric-field effects on optical absorption near thresholds in solids
Phys. Rev., 147 (1966), pp. 554–566
On the Ni(x) integral function and its application to the Airy’s non-homogeneous equation
Appl. Math. Comput., 217 (2011), pp. 7349–7360
Airy Functions and Applications to Physics
(second ed.)Imperial College Press, London (2010)
Copyright © 2013 Elsevier Inc. All rights reserved.