This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

\TitleHead

An elementary proof of the Voros connection formula for the Airy equation \dedicatoryDedicated to Professor Yoshitsugu Takei on his 60th birthday \AuthorHeadAoki, T., Suzuki, T. and Uchida, S. \supportThe first and the second author are supported by JSPS KAKENHI Grant No. 18K03385. \VolumeNox \YearNo202x \PagesNo000–000 \communicationReceived April 20, 202x. Revised September 11, 202x.

An elementary proof of the Voros connection formula for WKB solutions to the Airy equation with a large parameter

Takashi Aoki111Kindai University, Kowakae 3-4-1, Higashi-Osaka, 577-8502, Japan.
e-mail: [email protected]
   Takao Suzuki222Kindai University, Kowakae 3-4-1, Higashi-Osaka, 577-8502, Japan.
e-mail: [email protected]
 and Shofu Uchida333Kindai University, Kowakae 3-4-1, Higashi-Osaka, 577-8502, Japan.
e-mail: [email protected]
Abstract

The Voros connection formula for WKB solutions to the Airy equation with a large parameter is proved by using cubic equations. Some parts of the results are generalized to the Pearcey system, which is a two-variable version of the Airy equation, is given.

:
33C05, 34E05, 34E15, 34E20
keywords:
the Voros connection formula, Airy function, Pearcey integral, WKB solutions, resurgence

1 Introduction

In [3], the Voros connection formula (cf. [12]) for the WKB solutions to Schrödinger-type ordinary differential equations is proved from the viewpoint of microlocal analysis [6]. The proof consists of two parts. In the first part, the Schrödinger-type ordinary differential equation with an analytic potential is formally transformed to the Airy equation with a large parameter near a simple turning point. The formal series appearing in the transformation can be justified by using microdifferential operators. In the second part, the Voros connection formula for the Airy equation is proved by computing the Borel transform of the WKB solutions directly in terms of the Gauss hypergeometric functions. The classical connection formulas for the hypergeometric functions yield the Voros formula for the Airy equation. Combining these two parts, we have the Voros formula for general equations. (See [7] also.)

In this article, we focus on the second part. We observe that the parameters in the hypergeometric functions used in the proof in [3, 7] are contained in the Schwarz list [11]. In other words, they are algebraic functions. Our aim is to prove the connection formula for the WKB solutions to the Airy equation by using analytic continuation of algebraic functions, not of the hypergeometric functions. This point of view gives some generalization. We find a similar structure in the Pearcey system (cf. [9]) with a large parameter. This system is a natural generalization of the Airy equation to the two-variable case. We see that the Borel transforms of the suitably normalized WKB solutions to the Pearcey system are also algebraic functions. Hence such WKB solutions are resurgent. We hope that this example provides a part of basics of the prospective theory for the exact WKB analysis of holonomic systems.

2 The Airy equation with a large parameter and its WKB solutions

The differential equation

(1) (d2dz2+z)w=0\left(-\frac{d^{2}}{dz^{2}}+z\right)w=0

is called the Airy equation (cf. [10]). Airy used a solution to this equation expressed by an integral, known as the “Airy function” nowadays (see § 3), effectively in his theory of the rainbow (caustics) [1]. We introduce a positive large parameter η\eta by setting z=η2/3x,ψ(x,η)=w(η2/3x)z=\eta^{2/3}x,\,\psi(x,\eta)=w(\eta^{2/3}x). Then we have a differential equation of the form

(2) (d2dx2+η2x)ψ=0.\left(-\frac{d^{2}}{dx^{2}}+\eta^{2}x\right)\psi=0.

We call this the Airy equation with the large parameter, or simply, the Airy equation. This equation plays a fundamental role in the exact WKB analysis. Let SS denote the logarithmic derivative of the unknown function ψ\psi, namely S=ddxlogψ.\displaystyle S=\frac{d}{dx}\log\psi. Then SS should satisfy the following Riccati-type equation:

(3) dSdx+S2=η2x.\frac{dS}{dx}+S^{2}=\eta^{2}x.

This equation has a formal solution S=j=1ηjSj\displaystyle S=\sum_{j=-1}^{\infty}\eta^{-j}S_{j} defined by the recurrence relation

(4) {S12=x,Sj+1=12S1(dSjdx+k=0jSkSjk)(j=1,0,1,2,3,).\left\{\begin{split}\ \ S_{-1}^{2}&=x,\\ S_{j+1}&=-\frac{1}{2S_{-1}}\left(\frac{dS_{j}}{dx}+\sum_{k=0}^{j}S_{k}S_{j-k}\right)\quad(j=-1,0,1,2,3,\dots).\end{split}\right.

We consider the exponential of the integral of SS:

ψ=exp(S𝑑x),\psi=\exp\left(\int Sdx\right),

which formally satisfies (2). We call this a WKB solution to (2). We easily see that

S1=x1/2,S0=14x,S_{-1}=x^{1/2},\ S_{0}=-\frac{1}{4x},\,\dots\,

and if we fix the branch of the square root, say, as x1/2>0x^{1/2}>0 for x>0x>0, we have a formal solution S(+)S^{(+)}. Another choice of the branch also gives a formal solution S()S^{(-)}. If we set

Sodd=j=1η2j1S2j+1=ηx1/2η1532x5/2η311052048x11/2,S_{\rm odd}=\sum_{j=-1}^{\infty}\eta^{-2j-1}S_{2j+1}=\eta x^{1/2}-\eta^{-1}\frac{5}{32}x^{-5/2}-\eta^{-3}\frac{1105}{2048}x^{-11/2}-\cdots,
Seven=j=0η2jS2j=14xη21564x416951024x7+,S_{\rm even}=\sum_{j=0}^{\infty}\eta^{-2j}S_{2j}=-\frac{1}{4x}-\eta^{-2}\frac{15}{64}x^{-4}-\frac{1695}{1024}x^{-7}+\cdots,

then we have

S(±)=±Sodd+SevenS^{(\pm)}=\pm S_{\rm odd}+S_{\rm even}

satisfies (3) and

Seven=12ddxlogSoddS_{\rm even}=-\frac{1}{2}\frac{d}{dx}\log S_{\rm odd}

holds. Hence we may take 1/2logSodd-1/2\log S_{\rm odd} as a primitive of SevenS_{\rm even} and have the special WKB solutions of the form

(5) ψ±=1Soddexp(±0xSodd𝑑x).\psi_{\pm}=\frac{1}{\sqrt{S_{\rm odd}}}\exp\left(\pm\int_{0}^{x}S_{\rm odd}dx\right).

Here the integral is defined by one half of the term-by-term contour integral of SoddS_{\rm odd} starting from xx on the second sheet of the Riemann surface of x\sqrt{x}, going around the origin counterclockwise and back to the xx on the first sheet:

0xSodd𝑑x=η23x3/2+η1548x3/2+η311059216x9/2+.\int_{0}^{x}S_{\rm odd}dx=\eta\frac{2}{3}x^{3/2}+\eta^{-1}\frac{5}{48}x^{-3/2}+\eta^{-3}\frac{1105}{9216}x^{-9/2}+\cdots.

We observe that SoddS_{\rm odd} has the weighted homogeneity

Sodd(λ2x,λ3η)=λ2Sodd(x,η).S_{\rm odd}(\lambda^{2}x,\lambda^{-3}\eta)=\lambda^{-2}S_{\rm odd}(x,\eta).

Hence ψ±\psi_{\pm} satisfies

ψ±(λ2x,λ3η)=λψ±(x,η).\psi_{\pm}(\lambda^{2}x,\lambda^{-3}\eta)=\lambda\psi_{\pm}(x,\eta).

This relation and (2) imply that ψ±\psi_{\pm} are formal solutions to the following system of partial differential equations in the variable (x,η)(x,\eta):

(6) {(2x2+η2x)ψ=0,(2xx3ηη1)ψ=0.\left\{\begin{split}&\ \ \left(-\frac{\partial^{2}}{\partial x^{2}}+\eta^{2}x\right)\psi=0,\\ &\ \ \left(2x\frac{\partial}{\partial x}-3\eta\frac{\partial}{\partial\eta}-1\right)\psi=0.\end{split}\right.

3 Integral representation

There are two standard solutions to (1), which are called the Airy functions [10]:

(7) Ai(z)\displaystyle{\rm Ai}(z) =12πiγ1exp(zξ+ξ33)𝑑ξ,\displaystyle=\frac{1}{2\pi i}\int_{\gamma_{1}}\exp\left(-z\xi+\frac{\xi^{3}}{3}\right)d\xi,
(8) Bi(z)\displaystyle{\rm Bi}(z) =12πγ2+γ3exp(zξ+ξ33)𝑑ξ.\displaystyle=\frac{1}{2\pi}\int_{-\gamma_{2}+\gamma_{3}}\exp\left(-z\xi+\frac{\xi^{3}}{3}\right)d\xi.

Here the paths γj(j=1,2,3)\gamma_{j}\ (j=1,2,3) are taken as shown in Fig. 3.1. More precisely, the arguments of three half-line asymptotes of the paths are ±π/3,π\pm\pi/3,\pi.

[Uncaptioned image]

Figure 3.1.

γ2\gamma_{2}

γ1\gamma_{1}

γ3\gamma_{3}

Hence the integrals

(9) φj=γjexp(η(t33xt))𝑑t(j=1,2,3)\varphi_{j}=\int_{\gamma_{j}}\exp\left(\eta\left(\frac{t^{3}}{3}-xt\right)\right)dt\quad(j=1,2,3)

are solutions to (2) satisfying φ1+φ2+φ3=0\varphi_{1}+\varphi_{2}+\varphi_{3}=0. Here γj\gamma_{j}’s are suitably deformed. We can see that φj\varphi_{j} have the same homogeneity as that of ψ±\psi_{\pm}, namely,

φj(λ2x,λ3η)=λφj(x,η).\varphi_{j}(\lambda^{2}x,\lambda^{-3}\eta)=\lambda\varphi_{j}(x,\eta).

This implies ψ=φj\psi=\varphi_{j} (j=1,2,3j=1,2,3) satisfy (6).

Remark.

The Airy functions are expressed in terms of φj\varphi_{j} as

(10) Ai(η2/3x)=η1/32πiφ1(x,η),Bi(η2/3x)=η1/32π(φ3(x,η)φ2(x,η)).{\rm Ai}(\eta^{2/3}x)=\frac{\eta^{1/3}}{2\pi i}\varphi_{1}(x,\eta),\quad{\rm Bi}(\eta^{2/3}x)=\frac{\eta^{1/3}}{2\pi}(\varphi_{3}(x,\eta)-\varphi_{2}(x,\eta)).

We rewrite (9) by setting t3/3xt=yt^{3}/3-xt=-y:

(11) φj=cjg(x,y)exp(yη)𝑑y.\varphi_{j}=\int_{c_{j}}g(x,y)\exp(-y\eta)dy.

Here cjc_{j} is the image of γj\gamma_{j} by the mapping tyt\mapsto y and

g(x,y)=1xt2|t=t(x,y).g(x,y)=\left.\frac{1}{x-t^{2}}\right|_{t=t(x,y)}.

The branch of the root t=t(x,y)t=t(x,y) of the cubic equation

(12) t3/3xt=yt^{3}/3-xt=-y

is taken suitably (see §6). We note that (11) looks like the Laplace integral defining the Borel sum of WKB solutions (see (27)) and hence gg is expected to have some relation with the Borel transform of WKB solutions.

Lemma 3.1

The function g=g(x,y)g=g(x,y) defined as above satisfies the cubic equation

(13) (9y24x3)g3+3xg+1=0.(9y^{2}-4x^{3})g^{3}+3xg+1=0.
Proof.

Eliminating tt from the relations

(xt2)g=1,t33xt=y,(x-t^{2})g=1,\quad\frac{t^{3}}{3}-xt=-y,

we have (13). ∎

Proposition 3.2

The algebraic function gg defined by the cubic equation (13) satisfies the system of partial differential equations

(14) {(2x2+x2y2)g=0,(2xx+3yy+2)g=0.\left\{\begin{split}\ \ &\left(-\frac{\partial^{2}}{\partial x^{2}}+x\frac{\partial^{2}}{\partial y^{2}}\right)g=0,\\ \ \ &\left(2x\frac{\partial}{\partial x}+3y\frac{\partial}{\partial y}+2\right)g=0.\end{split}\right.
Proof.

We can express gxg_{x}, gxxg_{xx}, gyg_{y} and gyyg_{yy} in terms of gg by differentiating (13) twice in xx and in yy. Putting them into the left-hand sides of (14) and using (13), we see they vanish. ∎

We observe that the characteristic variety of the system (14) is the conormal bundle of the curve 4x39y2=04x^{3}-9y^{2}=0 and hence the system is holonomic. The second equation of (14) implies that the unknown function can be written in the form

g(x,y)=1xh(yx3/2)g(x,y)=\frac{1}{x}h\left(\frac{y}{x^{3/2}}\right)

by using a function hh of one variable. It is easy to find a second-order ordinary differential equation for hh by using the first equation. This implies the rank of the holonomic system equals 2. Hence we have

Theorem 3.3

Let gj(j=1,2,3)g_{j}\ (j=1,2,3) be three branches of the algebraic function gg defined by the cubic equation (13). Then any two of them form a basis of the analytic solution space of the holonomic system

(15) {(2x2+x2y2)u=0,(2xx+3yy+2)u=0.\left\{\ \begin{split}\ \ &\left(-\frac{\partial^{2}}{\partial x^{2}}+x\frac{\partial^{2}}{\partial y^{2}}\right)u=0,\\ \ \ &\left(2x\frac{\partial}{\partial x}+3y\frac{\partial}{\partial y}+2\right)u=0.\end{split}\right.

4 The Borel transform of WKB solutions

Let ψ±,B\psi_{\pm,B} denote the Borel transform of ψ±\psi_{\pm}. Explicitly, ψ±\psi_{\pm} have the forms

ψ±\displaystyle\psi_{\pm} =η12x14(1532η2x311052048η4x6)12\displaystyle=\eta^{-\frac{1}{2}}x^{-\frac{1}{4}}\left(1-\frac{5}{32}\eta^{-2}x^{-3}-\frac{1105}{2048}\eta^{-4}x^{-6}-\cdots\right)^{-\frac{1}{2}}
×exp(±(η23x32+η1548x32+η311059216x92+)).\displaystyle\hskip 28.45274pt\times\exp\left(\pm\left(\eta\frac{2}{3}x^{\frac{3}{2}}+\eta^{-1}\frac{5}{48}x^{-\frac{3}{2}}+\eta^{-3}\frac{1105}{9216}x^{-\frac{9}{2}}+\cdots\right)\right).

Setting

A=532x311052048η2x6A=-\frac{5}{32}x^{-3}-\frac{1105}{2048}\eta^{-2}x^{-6}-\cdots

and

B=548x32+η211059216x92+,B=\frac{5}{48}x^{-\frac{3}{2}}+\eta^{-2}\frac{1105}{9216}x^{-\frac{9}{2}}+\cdots,

we may rewrite them as

η12x14(1+η2A(x,η))12exp(±η23x32)exp(±η1B(x,η))\displaystyle\eta^{-\frac{1}{2}}x^{-\frac{1}{4}}(1+\eta^{-2}A(x,\eta))^{-\frac{1}{2}}\exp\left(\pm\eta\frac{2}{3}x^{\frac{3}{2}}\right)\exp(\pm\eta^{-1}B(x,\eta))
=η12x14exp(±η23x32)(112η2A38η4A2+)\displaystyle=\eta^{-\frac{1}{2}}x^{-\frac{1}{4}}\exp\left(\pm\eta\frac{2}{3}x^{\frac{3}{2}}\right)\left(1-\frac{1}{2}\eta^{-2}A-\frac{3}{8}\eta^{-4}A^{2}+\cdots\right)
×(1±η1B+η212B2±)\displaystyle\hskip 142.26378pt\times\left(1\pm\eta^{-1}B+\eta^{-2}\frac{1}{2}B^{2}\pm\cdots\right)
=η12x14exp(±η23x32)(1+b1±η1x32+b2±(η1x32)2+)\displaystyle=\eta^{-\frac{1}{2}}x^{-\frac{1}{4}}\exp\left(\pm\eta\frac{2}{3}x^{\frac{3}{2}}\right)(1+b_{1}^{\pm}\eta^{-1}x^{-\frac{3}{2}}+b_{2}^{\pm}(\eta^{-1}x^{-\frac{3}{2}})^{2}+\cdots)

with some constants bj±b_{j}^{\pm}. Then ψ±,B\psi_{\pm,B} can be written in the form

(16) ψ±,B(x,y)\displaystyle\psi_{\pm,B}(x,y) =1xπ(yx32±23)12{1+πb1±Γ(32)(yx32±23)+\displaystyle=\frac{1}{x\sqrt{\pi}}\left(\frac{y}{x^{\frac{3}{2}}}\pm\frac{2}{3}\right)^{-\frac{1}{2}}\left\{1+\frac{\sqrt{\pi}{b}_{1}^{\pm}}{\Gamma(\frac{3}{2})}\left(\frac{y}{x^{\frac{3}{2}}}\pm\frac{2}{3}\right)+\cdots\right.
+πbj±Γ(j+12)(yx32±23)j+}.\displaystyle\hskip 99.58464pt\cdots+\left.\frac{\sqrt{\pi}{b}_{j}^{\pm}}{\Gamma(j+\frac{1}{2})}\left(\frac{y}{x^{\frac{3}{2}}}\pm\frac{2}{3}\right)^{j}+\cdots\right\}.

We introduce a new variable s=3y4x32+12\displaystyle s=\frac{3y}{4x^{\frac{3}{2}}}+\frac{1}{2}. Then we may rewrite ψ±,B\psi_{\pm,B} as follows:

(17) ψ+,B\displaystyle\psi_{+,B} =32πxs12(1+b~1+43s+),\displaystyle=\frac{\sqrt{3}}{2\sqrt{\pi}x}s^{-\frac{1}{2}}\left(1+\tilde{b}_{1}^{+}\frac{4}{3}s+\cdots\right),
(18) ψ,B\displaystyle\psi_{-,B} =32πx(s1)12(1+b~143(s1)+).\displaystyle=\frac{\sqrt{3}}{2\sqrt{\pi}x}(s-1)^{-\frac{1}{2}}\left(1+\tilde{b}_{1}^{-}\frac{4}{3}(s-1)+\cdots\right).

Here we take the branches as s1/2>0s^{1/2}>0 if s>0s>0 and (s1)1/2>0(s-1)^{1/2}>0 if s>1s>1.

On the other hand, it follows from the definition of the Borel transform that ψ±,B\psi_{\pm,B} satisfies the formal Borel transform of (6):

(19) {(2x2+2y2x)ψ±,B=0(2xx3y(y)1)ψ±,B=0,\left\{\begin{split}\ &\left(-\frac{\partial^{2}}{\partial x^{2}}+\frac{\partial^{2}}{\partial y^{2}}x\right)\psi_{\pm,B}=0\\ &\left(2x\frac{\partial}{\partial x}-3\frac{\partial}{\partial y}(-y)-1\right)\psi_{\pm,B}=0,\end{split}\right.

or, equivalently,

(20) {(2x2+x2y2)ψ±,B=0,(2xx+3yy+2)ψ±,B=0.\left\{\ \begin{split}\ \ &\left(-\frac{\partial^{2}}{\partial x^{2}}+x\frac{\partial^{2}}{\partial y^{2}}\right)\psi_{\pm,B}=0,\\ \ \ &\left(2x\frac{\partial}{\partial x}+3y\frac{\partial}{\partial y}+2\right)\psi_{\pm,B}=0.\end{split}\right.

This system coincides with (15). Hence Theorem 3.3 yields

Proposition 4.1

The Borel transforms ψ±,B\psi_{\pm,B} of the WKB solutions ψ±\psi_{\pm} can be written as linear combinations of any two of gjg_{j}’s. Here gjg_{j} (j=1,2,3)(j=1,2,3) are defined in Theorem 3.3. Especially, ψ±,B\psi_{\pm,B} are algebraic functions.

Remark.

The explicit forms of ψ±,B\psi_{\pm,B} are given by using hypergeometric functions in [3, 7]:

(21) {ψ+,B=32πxs12F12(16,56,12;s),ψ,B=3i2πx(1s)12F12(56,16,12;1s).\left\{\begin{split}\quad\psi_{+,B}&=\frac{\sqrt{3}}{2\sqrt{\pi}\,x}s^{-\frac{1}{2}}{}_{2}F_{1}\left(\frac{1}{6},\frac{5}{6},\frac{1}{2};s\right),\\ \quad\psi_{-,B}&=\frac{\sqrt{3}\,i}{2\sqrt{\pi}\,x}(1-s)^{-\frac{1}{2}}{}_{2}F_{1}\left(\frac{5}{6},\frac{1}{6},\frac{1}{2};1-s\right).\end{split}\right.

The classical connection formulas for the hypergeometric functions were used in [3, 7] for the derivation of the Voros connection formula for the Airy equation. In the present article, we do not utilize (21).

We set

g=Xxs1/2(1s)1/2g=\frac{X}{x\,s^{1/2}(1-s)^{1/2}}

in (13). Then we have the following cubic equation for XX:

(22) 16X33Xs1/2(1s)1/2=0.16X^{3}-3X-s^{1/2}(1-s)^{1/2}=0.

If s=0s=0, we have X=0,±3/4X=0,\,\pm\!\sqrt{3}/4 and we find three roots X1,X2,X3X_{1},X_{2},X_{3} of (22) near s=0s=0 with expansions

(23) {X1=34+16s1/2163s+,X2=34+16s1/2+163s+,X3=13s1/25162s3/2.\left\{\begin{split}\ X_{1}&=\frac{\sqrt{3}}{4}+\frac{1}{6}s^{1/2}-\frac{1}{6\sqrt{3}}s+\cdots,\\ X_{2}&=-\frac{\sqrt{3}}{4}+\frac{1}{6}s^{1/2}+\frac{1}{6\sqrt{3}}s+\cdots,\\ X_{3}&=-\frac{1}{3}s^{1/2}-\frac{5}{162}s^{3/2}-\cdots.\end{split}\right.

We take the analytic continuation of XjX_{j} to s=1s=1. If s=1s=1, we also have X=0,±3/4X=0,\,\pm\!\sqrt{3}/4. For s=1/2s=1/2, we have a double root X=1/4X=-1/4 and a simple root X=1/2X=1/2. The expansion of the branches of XX which merge at s=1/2s=1/2 are

(24) X=14±123(s12)+118(s12)2±.X=-\frac{1}{4}\pm\frac{1}{2\sqrt{3}}\left(s-\frac{1}{2}\right)+\frac{1}{18}\left(s-\frac{1}{2}\right)^{2}\pm\cdots.

The graphs of Y=16X33Xs1/2(1s)1/2Y=16X^{3}-3X-s^{1/2}(1-s)^{1/2} for s=0, 1/2, 1s=0,\,1/2,\,1 are shown in Figures 4.1–4.3. Since the coefficients of (s1/2)(s-1/2) of (24) do not vanish, two roots X2X_{2} and X3X_{3} pass each other at s=1/2s=1/2 and the coefficients of the leading terms of the expansions of (X1,X2,X3)(X_{1},X_{2},X_{3}) at s=1s=1 are (3/4,0,3/4)(\sqrt{3}/4,0,-\sqrt{3}/4).

[Uncaptioned image][Uncaptioned image][Uncaptioned image]

34\displaystyle-\frac{\sqrt{3}}{4}                     34\displaystyle\frac{\sqrt{3}}{4}             X2=X3X_{2}=X_{3}               12\displaystyle\frac{1}{2}        34\displaystyle-\frac{\sqrt{3}}{4}                       34\displaystyle\frac{\sqrt{3}}{4}

X2X_{2}       X3X_{3}              X1X_{1}              14\displaystyle-\frac{1}{4}                     X1X_{1}         X3X_{3}       X2X_{2}             X1X_{1}

OO                                       OO                                       OO

Fig. 4.1. s=0s=0                   Fig. 4.2. s=1/2s=1/2                    Fig. 4.3. s=1s=1

Replacing ss by 1s1-s in the expansions of XjX_{j}’s at s=0s=0 and comparing the leading coefficients, we have

(25) {X1=34+16(1s)1/2163(1s)+,X2=13(1s)1/25162(1s)3/2,X3=34+16(1s)1/2+163(1s)+.\left\{\begin{split}\ X_{1}&=\frac{\sqrt{3}}{4}+\frac{1}{6}(1-s)^{1/2}-\frac{1}{6\sqrt{3}}(1-s)+\cdots,\\ X_{2}&=-\frac{1}{3}(1-s)^{1/2}-\frac{5}{162}(1-s)^{3/2}-\cdots,\\ X_{3}&=-\frac{\sqrt{3}}{4}+\frac{1}{6}(1-s)^{1/2}+\frac{1}{6\sqrt{3}}(1-s)+\cdots.\end{split}\right.

We set

gj=Xjxs1/2(1s)1/2(j=1,2,3).g_{j}=\frac{X_{j}}{xs^{1/2}(1-s)^{1/2}}\quad(j=1,2,3).

Then (23) yields the expansion of gjg_{j} at s=0s=0:

g1\displaystyle g_{1} =1xs1/2(34+16s1/2+5243s+),\displaystyle=\frac{1}{xs^{1/2}}\left(\frac{\sqrt{3}}{4}+\frac{1}{6}s^{1/2}+\frac{5}{24\sqrt{3}}s+\cdots\right),
g2\displaystyle g_{2} =1xs1/2(34+16s1/25243s+),g3=g1g2.\displaystyle=\frac{1}{xs^{1/2}}\left(-\frac{\sqrt{3}}{4}+\frac{1}{6}s^{1/2}-\frac{5}{24\sqrt{3}}s+\cdots\right),\quad g_{3}=-g_{1}-g_{2}.

Using Proposition 4.1 and comparing the leading terms of these expansions with (17), we have

(26) ψ+,B=1π(g1g2).\psi_{+,B}=\frac{1}{\sqrt{\pi}}(g_{1}-g_{2}).

Similarly, (25) yields the expansion of gjg_{j} at s=1s=1:

g1\displaystyle g_{1} =1x(1s)1/2(34+16(1s)1/2+5324(1s)+),\displaystyle=\frac{1}{x(1-s)^{1/2}}\left(\frac{\sqrt{3}}{4}+\frac{1}{6}(1-s)^{1/2}+\frac{5\sqrt{3}}{24}(1-s)+\cdots\right),
g2\displaystyle g_{2} =1x(13+1681(1s)+),\displaystyle=\frac{1}{x}\left(-\frac{1}{3}+\frac{16}{81}(1-s)+\cdots\right),
g3\displaystyle g_{3} =1x(1s)1/2(34+16(1s)1/25324(1s)+).\displaystyle=\frac{1}{x(1-s)^{1/2}}\left(-\frac{\sqrt{3}}{4}+\frac{1}{6}(1-s)^{1/2}-\frac{5\sqrt{3}}{24}(1-s)+\cdots\right).

We compare the leading terms with (18). Then we obtain ψ,B=i(g1g3)/π(=i(2g1+g2)/π)\psi_{-,B}={i}(g_{1}-g_{3})/{\sqrt{\pi}}\,(={i}(2g_{1}+g_{2})/{\sqrt{\pi}}). Thus we have

Theorem 4.2

The Borel transforms ψ±,B\psi_{\pm,B} of the WKB solutions ψ±\psi_{\pm} to the Airy equation are expressed in terms of three branches gjg_{j}’s ((specified as above)) of the algebraic function gg defined by (13) as follows:

ψ+,B=1π(g1g2),ψ,B=iπ(g1g3).\psi_{+,B}=\frac{1}{\sqrt{\pi}}(g_{1}-g_{2}),\quad\psi_{-,B}=\frac{i}{\sqrt{\pi}}(g_{1}-g_{3}).

These relations can be also written in the form

ψ+,B=1πΔ23x3/2g2(x,y),ψ,B=iπΔ23x3/2g3(x,y).\psi_{+,B}=\frac{1}{\sqrt{\pi}}\Delta_{-\frac{2}{3}x^{3/2}}g_{2}(x,y),\quad\psi_{-,B}=\frac{i}{\sqrt{\pi}}\Delta_{\frac{2}{3}x^{3/2}}g_{3}(x,y).

Here Δαg(x,y)\Delta_{\alpha}g(x,y) designates the discontinuity of g(x,y)g(x,y) at y=αy=\alpha.

Note that if x0x\neq 0, then ±2/3x3/2\pm 2/3x^{3/2} are singularities of square root-type of ψ+,B\psi_{+,B} and of ψ,B\psi_{-,B} in yy-variable.

5 The Voros connection formula for the Airy equation

The Stokes curve (cf. [7]) of (2) is defined by

Im0xx𝑑x=0.{\rm Im}\int_{0}^{x}\sqrt{x}\,dx=0.

It consists of three half-lines

argx=0,±23π.\arg x=0,\pm\frac{2}{3}\pi.

Let I{\mathcal{R}}_{\rm I} and II{\mathcal{R}}_{\rm II} denote the sectors (two of the Stokes regions)

{x|2/3π<argx<0}and{x| 0<argx<2/3π},\{\,x\in{\mathbb{C}}\,|\,-2/3\pi<\arg x<0\,\}\quad\mbox{and}\quad\{\,x\in{\mathbb{C}}\,|\,0<\arg x<2/3\pi\,\},

respectively. Let ±(x)\ell_{\pm}(x) be half-lines

{23x3/2+t|t0}\left\{\left.\,\mp\frac{2}{3}x^{3/2}+t\,\right|\,t\geq 0\,\right\}

with the positive orientation in the yy-plane and we set

(27) Ψ±J=±(x)ψ±,B(x,y)exp(yη)𝑑yforxJ(J=I,II).\Psi_{\pm}^{J}=\int_{\ell_{\pm}(x)}\psi_{\pm,B}(x,y)\exp(-y\eta)dy\quad\mbox{for}\ \ x\in{\mathcal{R}}_{J}\quad(J={\rm I,II}).

Figures 5.1 and 5.2 show J{\mathcal{R}}_{J} and ±(x)\ell_{\pm}(x) for xJx\in{\mathcal{R}}_{J} (J=I,IIJ={\rm I,II}), respectively. The wavy lines designate the branch cuts for ψ+,B\psi_{+,B} in yy-plane. Theorem 4.2 implies that these functions are well-defined and each Ψ±J\Psi_{\pm}^{J} gives the Borel sum of ψ±\psi_{\pm} in J{\mathcal{R}}_{J} (J=I,IIJ={\rm I,II}), respectively.

[Uncaptioned image][Uncaptioned image]
[Uncaptioned image]

23x3/2\frac{2}{3}x^{3/2}                                 23x3/2\frac{2}{3}x^{3/2}

(x)\ell_{-}(x)

23x3/2-\frac{2}{3}x^{3/2}                                23x3/2-\frac{2}{3}x^{3/2}

II{\mathcal{R}}_{\rm II}                                        +(x)\ell_{+}(x)

Fig. 5.2.1. xIIx\in{\mathcal{R}}_{\rm II}                  Fig. 5.2.2. xIIx\in{\mathcal{R}}_{\rm II}

OO

I{\mathcal{R}}_{\rm I}                                         +(x)\ell_{+}(x)

23x3/2-\frac{2}{3}x^{3/2}                                23x3/2-\frac{2}{3}x^{3/2}

23x3/2\frac{2}{3}x^{3/2}                                 23x3/2\frac{2}{3}x^{3/2}

(x)\ell_{-}(x)

Fig. 5.1. Stokes regions           Fig. 5.2.3. xIx\in{\mathcal{R}}_{\rm I}                   Fig. 5.2.4. xIx\in{\mathcal{R}}_{\rm I}

[Uncaptioned image]
[Uncaptioned image]

We take the analytic continuation of Ψ±I\Psi_{\pm}^{\rm I} to II{\mathcal{R}}_{\rm II}. If a point xIx\in{\mathcal{R}}_{\rm I} with Rex>0{\rm Re}\,x>0 moves up to II{\mathcal{R}}_{\rm II} across the positive real axis, the singular point y=2/3x3/2y=2/3x^{3/2} of ψ+,B\psi_{+,B} crosses +(x)\ell_{+}(x) (see Figures 5.2.3 and 5.2.1). Hence we have to take a path ~+(x)\tilde{\ell}_{+}(x) of integration shown in Figure 5.3 instead of +(x)\ell_{+}(x) for xIIx\in{\mathcal{R}}_{\rm II} for the analytic continuation.

[Uncaptioned image]
[Uncaptioned image]

~+(x)\tilde{\ell}_{+}(x)            23x3/2\frac{2}{3}x^{3/2}                                                 Γ\Gamma

​​​23x3/2-\frac{2}{3}x^{3/2}                                                                                       +(x)\ell_{+}(x)

Fig. 5.3.                                          Fig. 5.4.

Thus the analytic continuation of Ψ+I\Psi_{+}^{\rm I} has an expression

Ψ+I=~+(x)ψ+,B(x,y)exp(yη)𝑑y\Psi_{+}^{\rm I}=\int_{\tilde{\ell}_{+}(x)}\psi_{+,B}(x,y)\exp(-y\eta)dy

for xIIx\in{\mathcal{R}}_{\rm II}. If we take a path Γ\Gamma surrounding the branch cut (or (x)\ell_{-}(x)) clockwise (as shown in Figure 5.4), we have

~+(x)+(x)+Γ\tilde{\ell}_{+}(x)\sim\ell_{+}(x)+\Gamma

as the path of integration for xIIx\in{\mathcal{R}}_{\rm II}. Hence we have

(28) Ψ+I=+(x)ψ+,B(x,y)exp(yη)𝑑y+Γψ+,B(x,y)exp(yη)𝑑y.\Psi_{+}^{\rm I}=\int_{\ell_{+}(x)}\psi_{+,B}(x,y)\exp(-y\eta)dy+\int_{\Gamma}\psi_{+,B}(x,y)\exp(-y\eta)dy.

The first term equals Ψ+II\Psi_{+}^{\rm II}. It follows from Theorem 4.2 that the second term can be written in terms of gjg_{j}’s:

Γ1π(g1g2)exp(yη)𝑑y.\int_{\Gamma}\frac{1}{\sqrt{\pi}}(g_{1}-g_{2})\exp(-y\eta)dy.

Since g2g_{2} is holomorphic on a neighborhood of (x)\ell_{-}(x) in the yy-variable, it does not contribute to the value of the integration. Therefore we have

(29) Ψ+I=Ψ+II+1πΓg1exp(yη)𝑑y.\Psi_{+}^{\rm I}=\Psi_{+}^{\rm II}+\frac{1}{\sqrt{\pi}}\int_{\Gamma}g_{1}\exp(-y\eta)dy.

The second term of the right-hand side equals

1π(x)(Δ23x3/2g3)exp(yη)𝑑y.-\frac{1}{\sqrt{\pi}}\int_{\ell_{-}(x)}(\Delta_{\frac{2}{3}x^{3/2}}g_{3})\exp(-y\eta)dy.

Hence Theorem 4.2 yields

Ψ+I=Ψ+II1i(x)ψ,B(x,y)exp(yη)𝑑y,\Psi_{+}^{\rm I}=\Psi_{+}^{\rm II}-\frac{1}{i}\int_{\ell_{-}(x)}\psi_{-,B}(x,y)\exp(-y\eta)dy,

namely,

Ψ+I=Ψ+II+iΨII.\Psi_{+}^{\rm I}=\Psi_{+}^{\rm II}+i\Psi_{-}^{\rm II}.

On the other hand, the singularity y=2/3x3/2y=-2/3x^{3/2} does not meet (x)\ell_{-}(x) when xx moves from I{\mathcal{R}}_{\rm I} to II{\mathcal{R}}_{\rm II} across the positive real axis. This implies ΨI=ΨII\Psi_{-}^{\rm I}=\Psi_{-}^{\rm II}. Hence we have obtained the Voros connection formula for the WKB solutions to the Airy equation without using any knowledge concerning the hypergoemetric functions:

{Ψ+I=Ψ+II+iΨII,ΨI=ΨII.\left\{\begin{split}\ \Psi_{+}^{\rm I}&=\Psi_{+}^{\rm II}+i\,\Psi_{-}^{\rm II},\\ \Psi_{-}^{\rm I}&=\Psi_{-}^{\rm II}.\end{split}\right.

6 Relation between WKB solutions and Airy functions

In this section, we relate the Airy functions and the WKB solutions. We employ the same notation as in the previous sections.

Theorem 6.1

The Airy functions Ai{\rm Ai} and Bi{\rm Bi} are expressed in terms of the Borel sums Ψ±I\Psi_{\pm}^{\rm I} of WKB solutions ψ±\psi_{\pm} to the Airy equation in I{\mathcal{R}}_{\rm I} as follows:

(30) {Ai(η23x)=η1312πΨI,Bi(η23x)=η131πΨ+Iη13i2πΨI.\left\{\ \begin{split}{\rm Ai}(\eta^{\frac{2}{3}}x)&=\eta^{\frac{1}{3}}\frac{1}{2\sqrt{\pi}}\Psi_{-}^{\rm I},\\ {\rm Bi}(\eta^{\frac{2}{3}}x)&=\eta^{\frac{1}{3}}\frac{1}{\sqrt{\pi}}\Psi_{+}^{\rm I}-\eta^{\frac{1}{3}}\frac{i}{2\sqrt{\pi}}\Psi_{-}^{\rm I}.\end{split}\right.
Proof.

We choose the path γj\gamma_{j} of integration (9) more carefully. If xIx\in{\mathcal{R}}_{\rm I}, we take γ1\gamma_{1} (resp. γ2\gamma_{2}) as the steepest descent path of the phase function of the integral (9) passing through the saddle point x\sqrt{x} (resp. x-\sqrt{x}). Here we take the branch as |argx|<π/3|\arg\sqrt{x\,}\,|<\pi/3 for xIIIx\in{\mathcal{R}}_{\rm I}\cup{\mathcal{R}}_{\rm II}. As a set, γ1\gamma_{1} (resp. γ2\gamma_{2}) is included in

{t|Im(t33+xt+23x3/2)=0,Re(t33xt+23x3/2)0}\left\{\,t\in{\mathbb{C}}\,\left|\,{\rm Im}\left(\frac{t^{3}}{3}+xt+\frac{2}{3}x^{3/2}\right)=0,\,{\rm Re}\left(\frac{t^{3}}{3}-xt+\frac{2}{3}x^{3/2}\right)\leq 0\,\right.\right\}
(resp.{t|Im(t33xt23x3/2)=0,Re(t33xt23x3/2)0}).\left({\rm resp.}\left\{\,t\in{\mathbb{C}}\,\left|\,{\rm Im}\left(\frac{t^{3}}{3}-xt-\frac{2}{3}x^{3/2}\right)=0,\,{\rm Re}\left(\frac{t^{3}}{3}-xt-\frac{2}{3}x^{3/2}\right)\leq 0\,\right.\right\}\right).

We go back to (11) and specify the branch of gg more precisely. Let us divide γ1\gamma_{1} into two parts by cutting it at the saddle point x\sqrt{x}. We denote by γ1\gamma_{1}^{-} (resp. γ1+\gamma_{1}^{+}) the lower (resp. upper) part. Let c1±c_{1}^{\pm} be the image of γ1±\gamma_{1}^{\pm} by the mapping tyt\,\mapsto\,y. Then c1c_{1}^{-} (resp. c1+c_{1}^{+}) coincides with (x)-\ell_{-}(x) (resp. (x)\ell_{-}(x)). If yc1y\in c_{1}^{-} (resp. yc1+y\in c_{1}^{+}) is close to 2/3x3/22/3x^{3/2}, the branch of the root tt of (12) should be taken as

t=x1/2(1+233(1s)1/2+)(resp.t=x1/2(1233(1s)1/2+)).t=x^{1/2}\left(1+\frac{2\sqrt{3}}{3}(1-s)^{1/2}+\cdots\right)\quad\left(\mbox{resp.}\ \ t=x^{1/2}\left(1-\frac{2\sqrt{3}}{3}(1-s)^{1/2}+\cdots\right)\right).

Here we choose the branch as (1s)1/2=(s1)1/2eπi/2(1-s)^{1/2}=(s-1)^{1/2}e^{-\pi i/2} for s>1s>1. Hence the branch of g=1xt2\displaystyle g=\frac{1}{x-t^{2}} should be taken as g1g_{1} on c1c_{1}^{-} (resp. g3g_{3} on c1+c_{1}^{+}). By the definition of gjg_{j}, (11) for j=1j=1 should be understood as

φ1\displaystyle\varphi_{1} =c1g1(x,y)exp(yη)𝑑y+c1+g3(x,y)exp(yη)𝑑y\displaystyle=\int_{c_{1}^{-}}g_{1}(x,y)\exp(-y\eta)dy+\int_{c_{1}^{+}}g_{3}(x,y)\exp(-y\eta)dy
=(x)g1(x,y)exp(yη)𝑑y++(x)g3(x,y)exp(yη)𝑑y\displaystyle=\int_{-\ell_{-}(x)}g_{1}(x,y)\exp(-y\eta)dy+\int_{\ell_{+}(x)}g_{3}(x,y)\exp(-y\eta)dy
=(x)(g1(x,y)g3(x,y))exp(yη)𝑑y.\displaystyle=-\int_{\ell_{-}(x)}(g_{1}(x,y)-g_{3}(x,y))\exp(-y\eta)dy.

Using Theorem 4.2, we have

φ1=πi(x)ψ,B(x,y)exp(yη)𝑑y.\varphi_{1}=-\frac{\sqrt{\pi}}{i}\int_{\ell_{-}(x)}\psi_{-,B}(x,y)\exp(-y\eta)dy.

Thus the first equation of (10) shows

Ai(η2/3x)=η132πiπi(x)ψ,B(x,η)exp(yη)𝑑y=η1312πΨI.{\rm Ai}(\eta^{2/3}x)=-\frac{\eta^{\frac{1}{3}}}{2\pi i}\cdot\frac{\sqrt{\pi}}{i}\int_{\ell_{-}(x)}\psi_{-,B}(x,\eta)\exp(-y\eta)dy=\eta^{\frac{1}{3}}\frac{1}{2\sqrt{\pi}}\Psi_{-}^{\rm I}.

Next we divide γ2\gamma_{2} into two parts at the saddle point x-\sqrt{x}. We denote by γ2\gamma_{2}^{-} (resp. γ2+\gamma_{2}^{+}) the right (resp. left) part. Let c2c_{2}^{-} (resp. c2+c_{2}^{+}) be the image of γ2\gamma_{2}^{-} (resp. γ2+\gamma_{2}^{+}) by the mapping tyt\,\mapsto\,y. Then c2c_{2}^{-} (resp. c2+c_{2}^{+}) coincides with +(x)-\ell_{+}(x) (resp. +(x)\ell_{+}(x)). If yc2y\in c_{2}^{-} (resp. yc2+y\in c_{2}^{+}) is close to 2/3x3/2-2/3x^{3/2}, the branch of the root tt of (12) should be taken as

t=x1/2(1+233s1/2+)(resp.t=x1/2(1233s1/2+)).t=x^{1/2}\left(-1+\frac{2\sqrt{3}}{3}s^{1/2}+\cdots\right)\quad\left(\mbox{resp.}\ \ t=x^{1/2}\left(-1-\frac{2\sqrt{3}}{3}s^{1/2}+\cdots\right)\right).

Hence the branch of g=1xt2\displaystyle g=\frac{1}{x-t^{2}} must be taken as g1g_{1} on c1c_{1}^{-} (resp. g2g_{2} on c1+c_{1}^{+}). Then we have

φ2\displaystyle\varphi_{2} =c2g1(x,y)exp(yη)𝑑y+c2+g2(x,y)exp(yη)𝑑y\displaystyle=\int_{c_{2}^{-}}g_{1}(x,y)\exp(-y\eta)dy+\int_{c_{2}^{+}}g_{2}(x,y)\exp(-y\eta)dy
=+(x)(g1(x,y)g2(x,y))exp(yη)𝑑y\displaystyle=-\int_{\ell_{+}(x)}(g_{1}(x,y)-g_{2}(x,y))\exp(-y\eta)dy
=π+(x)ψ+,B(x,y)exp(yη)𝑑y.\displaystyle=-\sqrt{\pi}\int_{\ell_{+}(x)}\psi_{+,B}(x,y)\exp(-y\eta)dy.

Hence the second equation in (10) yields

Bi(η2/3x)\displaystyle{\rm Bi}(\eta^{2/3}x) =η132π(φ1(x,η)2φ2(x,η))\displaystyle=\frac{\eta^{\frac{1}{3}}}{2\pi}(-\varphi_{1}(x,\eta)-2\varphi_{2}(x,\eta))
=η132ππi(x)ψ,B(x,y)exp(yη)𝑑y\displaystyle=\frac{\eta^{\frac{1}{3}}}{2\pi}\frac{\sqrt{\pi}}{i}\int_{\ell_{-}(x)}\psi_{-,B}(x,y)\exp(-y\eta)dy
+η13π+(x)πψ+,B(x,y)exp(yη)𝑑y\displaystyle\hskip 71.13188pt+\frac{\eta^{\frac{1}{3}}}{\pi}\int_{\ell_{+}(x)}\sqrt{\pi}\psi_{+,B}(x,y)\exp(-y\eta)dy
=η13πΨ+Iη132πiΨI.\displaystyle=\frac{\eta^{\frac{1}{3}}}{\sqrt{\pi}}\Psi_{+}^{\rm I}-\frac{\eta^{\frac{1}{3}}}{2\sqrt{\pi}}i\Psi_{-}^{\rm I}.

This completes the proof. ∎

Remark.

Using the Voros connection formula, we obtain

{Ai(η23x)=η1312πΨII,Bi(η23x)=η131πΨ+II+η13i2πΨII.\left\{\ \begin{split}{\rm Ai}(\eta^{\frac{2}{3}}x)&=\eta^{\frac{1}{3}}\frac{1}{2\sqrt{\pi}}\Psi_{-}^{\rm II},\\ {\rm Bi}(\eta^{\frac{2}{3}}x)&=\eta^{\frac{1}{3}}\frac{1}{\sqrt{\pi}}\Psi_{+}^{\rm II}+\eta^{\frac{1}{3}}\frac{i}{2\sqrt{\pi}}\Psi_{-}^{\rm II}.\end{split}\right.

Conversely, Ψ±J\Psi_{\pm}^{\rm J} are written in terms of Ai{\rm Ai} and Bi{\rm Bi} as follows:

{Ψ+I=η13πiAi(η23x)+η13πBi(η23x),ΨI=η132πAi(η23x),\left\{\ \begin{split}\Psi_{+}^{\rm I}&=\eta^{-\frac{1}{3}}\sqrt{\pi}\,i\,{\rm Ai}(\eta^{\frac{2}{3}}x)+\eta^{-\frac{1}{3}}\sqrt{\pi}\,{\rm Bi}(\eta^{\frac{2}{3}}x),\\ \Psi_{-}^{\rm I}&=\eta^{-\frac{1}{3}}2\sqrt{\pi}\,{\rm Ai}(\eta^{\frac{2}{3}}x),\end{split}\right.
{Ψ+II=η13πiAi(η23x)+η13πBi(η23x),ΨII=η132πAi(η23x).\left\{\ \begin{split}\Psi_{+}^{\rm II}&=-\eta^{-\frac{1}{3}}\sqrt{\pi}\,i\,{\rm Ai}(\eta^{\frac{2}{3}}x)+\eta^{-\frac{1}{3}}\sqrt{\pi}\,{\rm Bi}(\eta^{\frac{2}{3}}x),\\ \Psi_{-}^{\rm II}&=\eta^{-\frac{1}{3}}2\sqrt{\pi}\,{\rm Ai}(\eta^{\frac{2}{3}}x).\end{split}\right.

Watson’s lemma and the expressions given above reproduce the classical asymptotic formulas for the Airy functions (cf. [10]):

{Ai(η23x)η1312πψ,Bi(η23x)η131πψ+η13i2πψ(xI,η),\left\{\ \begin{split}{\rm Ai}(\eta^{\frac{2}{3}}x)&\sim\eta^{\frac{1}{3}}\frac{1}{2\sqrt{\pi}}\psi_{-},\\ {\rm Bi}(\eta^{\frac{2}{3}}x)&\sim\eta^{\frac{1}{3}}\frac{1}{\sqrt{\pi}}\psi_{+}-\eta^{\frac{1}{3}}\frac{i}{2\sqrt{\pi}}\psi_{-}\end{split}\right.\quad(x\in{\mathcal{R}}_{\rm I},\ \eta\rightarrow\infty),
{Ai(η23x)η1312πψ,Bi(η23x)η131πψ++η13i2πψ(xII,η).\left\{\ \begin{split}{\rm Ai}(\eta^{\frac{2}{3}}x)&\sim\eta^{\frac{1}{3}}\frac{1}{2\sqrt{\pi}}\psi_{-},\\ {\rm Bi}(\eta^{\frac{2}{3}}x)&\sim\eta^{\frac{1}{3}}\frac{1}{\sqrt{\pi}}\psi_{+}+\eta^{\frac{1}{3}}\frac{i}{2\sqrt{\pi}}\psi_{-}\end{split}\right.\quad(x\in{\mathcal{R}}_{\rm II},\ \eta\rightarrow\infty).

Explicit forms of ψ±\psi_{\pm} are written as follows:

ψ±=e±23x3/2η2πn=0ηn12(±34)nΓ(n+16)Γ(n+56)n!x32n14.\psi_{\pm}=\frac{{e^{\pm\frac{2}{3}x^{3/2}\eta}}}{2\pi}\sum_{n=0}^{\infty}{\eta^{-n-\frac{1}{2}}}\left(\pm\frac{3}{4}\right)^{n}\frac{\Gamma\big{(}n+\frac{1}{6}\big{)}\Gamma\big{(}n+\frac{5}{6}\big{)}}{n!}x^{-\frac{3}{2}n-\frac{1}{4}}.

7 Some generalization

We consider the following integral:

(31) v=exp(η(t4+x2t2+x1t))𝑑t.v=\int\exp\left(\eta\left(t^{4}+x_{2}t^{2}+x_{1}t\right)\right)dt.

Here x1,x2x_{1},x_{2} are complex variables and the path of integration is taken suitably. This is a natural extension of (9) to two-variable case. It is called the Pearcey integral ([10]) with the large parameter η\eta. Most parts of the discussions developed in §§2–6 can be generalized to the system of partial differential equations that characterizes this integral. We review some results concerning this system without proof. Details will be given in [4].

We can easily see that ψ=v\psi=v satisfies the system of partial differential equations

(32) {(413+2x2η21+x1η3)ψ=0,(η212)ψ=0,\left\{\ \ \begin{split}&\left(4\partial_{1}^{3}+2x_{2}\eta^{2}\partial_{1}+x_{1}\eta^{3}\right)\psi=0,\\ &\left(\eta\partial_{2}-\partial_{1}^{2}\right)\psi=0,\end{split}\right.

where we set 1=/x1\partial_{1}=\partial/\partial x_{1}, 2=/x2\partial_{2}=\partial/\partial x_{2}. WKB solutions to this system is constructed in [2, 5] and the connection problem of the solutions was discussed in [5]. We employ another system of partial differential equations. We set

(33) P1\displaystyle{P}_{1} =412+2ηx21+η2x1,\displaystyle=4\partial_{1}\partial_{2}+2\eta x_{2}\partial_{1}+\eta^{2}x_{1},
(34) P2\displaystyle{P}_{2} =422+ηx11+2ηx22+η,\displaystyle=4\partial_{2}^{2}+\eta x_{1}\partial_{1}+2\eta x_{2}\partial_{2}+\eta,
(35) P3\displaystyle{P}_{3} =η212.\displaystyle=\eta\partial_{2}-\partial_{1}^{2}.

It is easy to see that ψ=v\psi=v is a solution to the system (cf. [9])

(36) {P1ψ=0,P2ψ=0,P3ψ=0.\left\{\begin{split}P_{1}\psi&=0,\\ P_{2}\psi&=0,\\ P_{3}\psi&=0.\end{split}\right.

In fact, we have

P1v=ηtexp(η(t4+x2t2+x1t))dt=0,P_{1}v=\eta\int\partial_{t}\exp\left(\eta\left(t^{4}+x_{2}t^{2}+x_{1}t\right)\right)dt=0,
P2v=ηt((texp(η(t4+x2t2+x1t)))dt=0,P_{2}v=\eta\int\partial_{t}(\left(t\exp\left(\eta\left(t^{4}+x_{2}t^{2}+x_{1}t\right)\right)\right)dt=0,
P3v=η2(t2t2)exp(η(t4+x2t2+x1t))𝑑t=0.P_{3}v=\eta^{2}\int\left(t^{2}-t^{2}\right)\exp\left(\eta\left(t^{4}+x_{2}t^{2}+x_{1}t\right)\right)dt=0.

Setting

Q1=413+2x2η21+x1η3andQ2=η212(=P3),Q_{1}=4\partial_{1}^{3}+2x_{2}\eta^{2}\partial_{1}+x_{1}\eta^{3}\quad\mbox{and}\quad Q_{2}=\eta\partial_{2}-\partial_{1}^{2}(=P_{3}),

we can confirm the following relations:

(37) P1\displaystyle P_{1} =η1(Q1+41Q2),\displaystyle=\eta^{-1}(Q_{1}+4\partial_{1}Q_{2}),
(38) P2\displaystyle P_{2} =η21Q1+(4η2Q2+8η212+2x2)Q2,\displaystyle=\eta^{-2}\partial_{1}Q_{1}+(4\eta^{-2}Q_{2}+8\eta^{-2}\partial_{1}^{2}+2x_{2})Q_{2},
(39) Q1\displaystyle Q_{1} =ηP141P3.\displaystyle=\eta P_{1}-4\partial_{1}P_{3}.

Hence if η0\eta\neq 0 is fixed, (36) is equivalent to (32). We also note that

P2=η11P1+2(2η12+x2)P3P_{2}=\eta^{-1}\partial_{1}P_{1}+2(2\eta^{-1}\partial_{2}+x_{2})P_{3}

holds. Next we consider η\eta as an independent complex variable. Then the systems (32) and (36) are subholonomic. To find another independent differential equation for vv, we look at the weighted homogeneity of (31) in (x1,x2,η)(x_{1},x_{2},\eta). We can see that

v(λ3x1,λ2x2,λ4η)=λv(x1,x2,η)v(\lambda^{3}x_{1},\lambda^{2}x_{2},\lambda^{-4}\eta)=\lambda v(x_{1},x_{2},\eta)

holds for λ0\lambda\neq 0 and hence ψ=v\psi=v is a solution to

(40) (3x11+2x224ηη1)ψ=0,(3x_{1}\partial_{1}+2x_{2}\partial_{2}-4\eta\partial_{\eta}-1)\psi=0,

where η=/η\partial_{\eta}=\partial/\partial\eta. We set

P4=3x11+2x224ηη1P_{4}=3x_{1}\partial_{1}+2x_{2}\partial_{2}-4\eta\partial_{\eta}-1

and consider the system of partial differential equations

Pjψ=0(j=1,2,3,4).P_{j}\psi=0\quad(j=1,2,3,4).

To be more specific, let 𝒟\mathcal{D} denote the Weyl algebra of the variable (x1,x2,η)(x_{1},x_{2},\eta) and \mathcal{I} the left ideal in 𝒟\mathcal{D} generated by PjP_{j} (j=1,2,3,4j=1,2,3,4). We can prove the following theorem (cf. [8]):

Theorem 7.1

Let {\mathcal{M}} denote the left 𝒟\mathcal{D}-module defined by :{\mathcal{I}}:

=𝒟/.{\mathcal{M}}={\mathcal{D}}/{\mathcal{I}}.

Then {\mathcal{M}} is a holonomic system of rank 33.

Thus {\mathcal{M}} characterizes the 3-dimensional linear subspace spanned by (31) in the space of analytic functions. Note that there are four valleys of the integral (31) and hence six infinite paths of integration connecting distinct two valleys. Any three of them are independent, which give a basis of the solution space.

Next we construct WKB solutions to {\mathcal{M}}. We set S=1ψ/ψS=\partial_{1}\psi/\psi and T=2ψ/ψT=\partial_{2}\psi/\psi. Since we have (37) and (38), we can use (32) to find S,TS,T. That is, we see that SS and TT should satisfy

4S3+2η2x2S+η3x1+12S1S+412S=04S^{3}+2\eta^{2}x_{2}S+\eta^{3}x_{1}+12S\partial_{1}S+4\partial_{1}^{2}S=0

and

ηT1SS2=0.\eta T-\partial_{1}S-S^{2}=0.

We seek formal solutions to these equations of the forms

S=k=1ηkSk,T=k=1ηkTk.S=\sum_{k=-1}^{\infty}\eta^{-k}S_{k},\quad T=\sum_{k=-1}^{\infty}\eta^{-k}T_{k}.

Putting these expressions into the above equations, we see that SkS_{k} and TkT_{k} are obtained by the following recursion relations and initial conditions:

(41) 4\displaystyle 4 S13+2x2S1+x1=0,\displaystyle S_{-1}^{3}+2x_{2}S_{-1}+x_{1}=0,
(42) S0\displaystyle S_{0} =121log(6S12+x2),\displaystyle=-\frac{1}{2}\partial_{1}\log(6S_{-1}^{2}+x_{2}),
(43) Sk\displaystyle S_{k} =26S12+x2(k1+k2+k3=k21k1,k2,k3<kSk1Sk2Sk3\displaystyle=-\frac{2}{6S_{-1}^{2}+x_{2}}\left(\sum_{\underset{-1\leq k_{1},\,k_{2},\,k_{3}<k}{k_{1}+k_{2}+k_{3}=k-2}}S_{k_{1}}S_{k_{2}}S_{k_{3}}\right.
+ 3k1+k2=k21k1,k2<kSk11Sk2+12Sk2)(k1),\displaystyle\hskip 113.81102pt\left.+\,3\sum_{\underset{-1\leq k_{1},\,k_{2}<k}{k_{1}+\,k_{2}=k-2}}S_{k_{1}}\partial_{1}S_{k_{2}}+\partial_{1}^{2}S_{k-2}\right)\ \ (k\geq 1),
(44) T1\displaystyle T_{-1} =S12,\displaystyle=S_{-1}^{2},
(45) Tk\displaystyle T_{k} =1Sk1+j=1kSjSkj1(k0).\displaystyle=\partial_{1}S_{k-1}+\sum_{j=-1}^{k}S_{j}S_{k-j-1}\ \ (k\geq 0).

This construction is the same as that given in [2, 5] and hence the 1-form of formal series

ω=Sdx1+Tdx2\omega=Sdx_{1}+Tdx_{2}

is closed. In these references, a formal solution of the form

ψ=η1/2exp((a1,a2)(x1,x2)ω)\psi=\eta^{-1/2}\exp\left(\int_{(a_{1},a_{2})}^{(x_{1},x_{2})}\omega\right)

is called a WKB solution to (32). Here (a1,a2)(a_{1},a_{2}) is a suitably fixed point.

Now we consider the WKB solutions to {\mathcal{M}} of the form

ψ=η1/2exp(ω).\psi=\eta^{-1/2}\exp\left(\int\omega\right).

In addition to (36) (or (32)), ψ\psi should satisfy (40) and hence the choice of the primitive of ω=k=1ηkωk=k=1ηk(Skdx1+Tkdx2)\displaystyle\omega=\sum_{k=-1}^{\infty}\eta^{-k}\omega_{k}=\sum_{k=-1}^{\infty}\eta^{-k}(S_{k}dx_{1}+T_{k}dx_{2}) is constrained by this equation (up to genuine additive constants), namely,

(46) {ω0=12log(6S12+x2),ωk=14k(3x1Sk+2x2Tk)(k0).\left\{\begin{split}\int\omega_{0}&=-\frac{1}{2}\log(6S_{-1}^{2}+x_{2}),\\ \int\omega_{k}&=-\frac{1}{4k}(3x_{1}S_{k}+2x_{2}T_{k})\quad(k\neq 0).\end{split}\right.

This choice is consistent with the construction of SkS_{k} and TkT_{k}. In fact, we can confirm the first equation of (46) by direct computation and the second by using the homogeneity of SkS_{k} and TkT_{k}. From now on, we take special WKB solutions of the form

(47) ψ=1(η(6S12+x2))1/2exp(ηω1+k=1ηkωk)\psi=\frac{1}{\left(\eta(6S_{-1}^{2}+x_{2})\right)^{1/2}}\exp\left(\eta\int\omega_{-1}+\sum_{k=1}^{\infty}\eta^{-k}\int\omega_{k}\right)

with the primitives given by (46). Let S1(j)S_{-1}^{(j)} (j=1,2,3j=1,2,3) denote the three roots of (41) and set T1(j)=(S1(j))2T_{-1}^{(j)}=(S_{-1}^{(j)})^{2}. According to this choice, we have three WKB solutions ψj\psi_{j} (j=1,2,3j=1,2,3) of the form (47).

Let ψj,B\psi_{j,B} be the Borel transform of ψj\psi_{j} (j=1,2,3j=1,2,3) and Pk,BP_{k,B} the formal Borel transform of PkP_{k} (k=1,2,3,4k=1,2,3,4) . The explicit forms of Pk,BP_{k,B}’s are given as follows:

P1,B\displaystyle P_{1,B} =412+2x2y1+x1y2,\displaystyle=4\partial_{1}\partial_{2}+2x_{2}\partial_{y}\partial_{1}+x_{1}\partial_{y}^{2},
P2,B\displaystyle P_{2,B} =422+x1y1+2x2y2+y,\displaystyle=4\partial_{2}^{2}+x_{1}\partial_{y}\partial_{1}+2x_{2}\partial_{y}\partial_{2}+\partial_{y},
P3,B\displaystyle P_{3,B} =y212,\displaystyle=\partial_{y}\partial_{2}-\partial_{1}^{2},
P4,B\displaystyle P_{{4,B}} =3x11+2x224y(y)1\displaystyle=3x_{1}\partial_{1}+2x_{2}\partial_{2}-4\partial_{y}(-y)-1
(\displaystyle( =3x11+2x22+4yy+3).\displaystyle=3x_{1}\partial_{1}+2x_{2}\partial_{2}+4y\partial_{y}+3).

Then Pk,Bψj,B=0P_{k,B}\psi_{j,B}=0 holds for j=1,2,3;k=1,2,3,4j=1,2,3;k=1,2,3,4. We denote by B{\mathcal{I}}_{B} the left ideal in 𝒟B{\mathcal{D}}_{B} generated by Pk,BP_{k,B} (k=1,2,3,4k=1,2,3,4). Here 𝒟B{\mathcal{D}}_{B} denotes the Weyl algebra of the variable (x1,x2,y)(x_{1},x_{2},y). Then the following theorem can be proved in a similar manner to Theorem 7.1

Theorem 7.2

Let B{\mathcal{M}}_{B} denote the left 𝒟B{\mathcal{D}}_{B}-module defined by B:{\mathcal{I}}_{B}:

B=𝒟B/B.{\mathcal{M}}_{B}={\mathcal{D}}_{B}/{\mathcal{I}}_{B}.

Then B{\mathcal{M}}_{B} is a holonomic system of rank 33.

Thus B{\mathcal{M}}_{B} characterizes the subspace of analytic functions spanned by ψj,B\psi_{j,B} (j=1,2,3j=1,2,3).

We go back to (31). We set t4+x2t2+x1t=yt^{4}+x_{2}t^{2}+x_{1}t=-y and rewrite (31) as

v=exp(ηy)g(x1,x2,y)𝑑y.v=\int\exp(-\eta y)g(x_{1},x_{2},y)dy.

Here gg is defined by

g(x1,x2,y)=14t3+2x2t+x1|t=t(x1,x2)g(x_{1},x_{2},y)=\left.\frac{1}{4t^{3}+2x_{2}t+x_{1}}\right|_{t=t(x_{1},x_{2})}

and the path of integration is suitably modified. The following lemma can be proved in a similar way to the proof of Lemma 3.1.

Lemma 7.3

The function gg defined as above satisfies the quadratic equation

(4x12x2(36yx22)+16y(x224y)227x14)g4+2(8x2y+2x23+9x12)g28x1g+1=0(4x_{1}^{2}x_{2}(36y-x_{2}^{2})+16y(x_{2}^{2}-4y)^{2}-27x_{1}^{4})\,g^{4}+2(-8x_{2}y+2x_{2}^{3}+9x_{1}^{2})\,g^{2}-8x_{1}g+1=0

and it is a solution to the holonomic system B{\mathcal{M}}_{B}.

We note that the singular locus of B{\mathcal{M}}_{B} coincides with the zero-point set of the leading coefficient of the above quadratic equation. For general (x1,x2,y)(x_{1},x_{2},y), there are four roots gkg_{k} (k=1,2,3,4)(k=1,2,3,4) of the quadratic equation, which satisfy g1+g2+g3+g4=0g_{1}+g_{2}+g_{3}+g_{4}=0. Looking at the singularity of gkg_{k}, we find that any three of gkg_{k}’s are linearly independent. Thus we have the following theorem.

Theorem 7.4

The Borel transform ψj,B\psi_{j,B} of the WKB solution ψj\psi_{j} (j=1,2,3)(j=1,2,3) can be written as a linear combination of any three of gkg_{k}’s. In particular, ψj,B\psi_{j,B}’s are algebraic and hence they are resurgent.

We can write down ψj,B\psi_{j,B} in terms of gkg_{k}’s. Explicit forms will be given in [4].

References

  • [1] Airy, G. B., On the intensity of light in the neighbourhood of a caustic, Transactions of the Cambridge Philosophical Society, 6, 3 (1838), 379–402.
  • [2] Aoki, T., Toward the exact WKB analysis of holonomic systems, RIMS Kôkyûroku 1433 (2005), 1–8 (in Japanese).
  • [3] Aoki, T., Kawai, T. and Takei,Y., The Bender-Wu analysis and the Voros theory, Special Functions, ICM-90 Satellite Conference Proceedings, Springer, 1991, 1–29.
  • [4] Aoki, T., Suzuki, T. and Uchida, S., The Borel transform of the WKB solutions to the Pearcey system, in preparation.
  • [5] Hirose, S., On the Stokes geometry for the Pearcey system and the (1,4) hypergeometric system, RIMS Kôkyûroku Bessatsu 40 (2013), 243–292.
  • [6] Kashiwara, M., Kawai, T. and Kimura, T., Foundation of Algebraic Analysis, PMS-37, Princeton University Press, 1986.
  • [7] Kawai, T. and Takei, Y., Algebraic Analysis of Singular Perturbation Theory, Translation of Mathematical Monographs, vol. 227, AMS, 2005.
  • [8] Oaku, T., Computation of the characteristic variety and the singular locus of a system of differential equations with polynomial coefficients. Japan J. Indust. Appl. Math., 11 (1994), 485-497.
  • [9] Okamoto, K. and Kimura, H., On particular solutions of the Garnier systems and the hypergeometric functions of several variables, Quart. J. Math., 37 (1986), 61–80.
  • [10] Olver, F. W. J. et al., NIST Handbook of Mathematical Functions, Cambridge University Press (2010).
  • [11] Schwarz, H. A., Ueber diejenigen Fälle in welchen die Gaussische hypergeometrische Reihe eine algebraische Function ihres vierten Elementes darstellt, Journal für die reine und angewandte Mathematik, 75 (1873), 292–335.
  • [12] Voros, A., The return of the quartic oscillator. The complex WKB method, Annales de l’I. H. P., section A, 39, no 3 (1983), p. 211-338.