Two Lax systems for the Painlev´e II equation, and two related kernels in random matrix theory Karl Liechty Dong Wang ∗ † 6 August 22, 2016 1 0 2 g u Abstract A We consider two Lax systems for the homogeneous Painlev´e II equation: one of size 2 2 9 studiedbyFlaschkaandNewellintheearly1980’s,andoneofsize4 4introducedbyDelvau×x– 1 × Kuijlaars–Zhang and Duits–Geudens in the early 2010’s. We prove that solutions to the 4 4 × systemcanbederivedfromthosetothe2 2systemviaanintegraltransform,andconsequently ] × I relatetheStokesmultipliersforthetwosystems. Ascorollariesweareabletoexpresstwokernels S for determinantalprocessesas contour integralsinvolvingthe Flaschka–NewellLax system: the . n tacnode kernel arising in models of nonintersecting paths, and a critical kernel arising in a i l two-matrix model. n [ 2 1 Introduction and statement of results v 3 The homogeneous Painlev´e II equation (PII) is the second order nonlinear ODE 0 6 1 y′′ = xy+2y3. (1.1) 0 . 1 Despite its unassuming form, its solutions, known as the Painlev´e transcendents, appear in exact 0 solutions of many models in mathematical physics. For example, one particular solution to (1.1) is 6 the one satisfying the boundary condition 1 : v q(σ) Ai(σ) as σ + , (1.2) Xi ∼ → ∞ r where Ai is the Airy function. This solution is known as the Hastings–McLeod solution [21]. It a is particularly important in random matrix theory, for it defines the celebrated Tracy–Widom distributions which describe the generic soft edge behavior of random matrices from orthogonal-, unitary-, or symplectic-invariant ensembles [29], [30]. The PII equation (1.1) is an integrable equation, and its integrability is characterized by the existence of Lax pairs. A Lax pair, or more generally a Lax system, is a system of overdetermined ∗Department of Mathematical Sciences, DePaul University,Chicago, IL,[email protected] Supported by DePaul University College of Science and Health Summer Research Grant, an AMS–Simons travel grant, and a grant from theSimons Foundation (#357872, Karl Liechty) †Department of Mathematics, National Universityof Singapore, Singapore, 119076, [email protected] Supported partially by thestartup grant R-146-000-164-133 1 lineardifferentialequationswhosecompatibilityimpliesanonlinearequation. LetΨ = Ψ(z ,...,z ) 1 r be an n n matrix-valued function with variables z ,...,z . Let 1 r × ∂Ψ ∂Ψ = A Ψ, ..., = A Ψ, (1.3) 1 r ∂z ∂z 1 r bean(overdetermined)systemofdifferentialequationssatisfiedbyΨwithn ncoefficientmatrices × A ,...,A . For the overdetermined differential equations to have nontrivial solutions, we need the 1 r compatibility among A ,...,A , the Frobenius compatibility conditions, sometimes called zero- 1 r curvature relations: ∂A ∂A i j +[A ,A ] =0, for all i,j = 1,...,r. (1.4) i j ∂z − ∂z j i TheFrobeniuscompatibility conditionsareingeneralnonlineardifferentialequationsfortheentries of A , and we call the system (1.3) the Lax system for the nonlinear equation(s) (1.4). In the most j common cases r = 2 and we call the system (1.3) a Lax pair, but we may also consider the general case r 2. ≥ Remark 1.1. The term Lax pair originates with the work of Peter Lax in the late 1960’s [27], in which he used thecompatibility of a pair of linear differential equations to study a nonlinear partial differential equation. In the problem considered by Lax the evolution of the time variable gives an isospectral deformation of the linear operator. On the other hand, Painlev´e equations represent isomonodromic deformations of theanalogous linear equations withrespecttothesingularities, i.e., the monodromy data is invariant as the argument of the (fixed) Painlev´e function changes, and the isomonodromic relations are expressed in the same form of Lax pairs [17, Chapter 4]. The idea of representing the Painlev´e equations as isomonodromy deformations of a system of linear equations is nearly as old as the Painlev´e equations themselves, dating back to the work of Fuchs [18] and later Garnier [19]. Therefore it may be more appropriate to call the overdetermined systems (1.3) and (1.18) Garnier–Fuchs pairs/systems rather than Lax pairs/systems. Such terminology can be found in the literature, see [24] and [25]. However, the phrase Lax pair is much more abundant in the literature and this is the nomenclature we use, following the terminology of [12], [14], [10], and [17]. Nonlinear differential equations which possess a Lax system representation are in some sense integrable, although they can be rather complicated. All of the Painlev´e equations, including (1.1), can be represented by Lax pairs/systems [17]. However, the construction of Lax pairs/systems for a given Painlev´e equation is far from trivial, and the relations between different Lax pair/systems for a Painlev´e equation deserve investigation for their own sake. In this paper we demonstrate the relation between one classical Lax pair and a recently discovered Lax system for the PII equation (1.1). However, the main motivation of our paper is not purely theoretical, but is driven by the appearance of these Lax systems in random matrix theory and related problems. The classical Lax pair and the new Lax system are both related to random matrix theory, but in quite different aspects. 1.1 The Flaschka–Newell Lax pair for PII First we present a classical Lax pair for (1.1), found by Flaschka and Newell [16]. Remark 1.2. The Flaschka–Newell Lax pair was originally presented for the general PII equation which has a free parameter (see Section 1.5), and we only present it for the homogeneous case 2 (1.1). A different Lax pair for PII was found by Jimbo and Miwa around the same time [22] (with a precursor in [19]), but in the homogeneous case the Jimbo–Miwa Lax pair can be reduced to the Flaschka–Newell one [17, Section 4.2]. Other Lax pairs associated to PII have been found by Harnad, Tracy, and Widom in [20] (of size 2 2) and by Joshi, Kitaev, and Treharne in [25] (of × size 3 3). The equivalence among these Lax pairs is discussed in [25]. × Let Φ = Φ(ζ;σ) be a 2 2 matrix-valued function with variables ζ and σ which satisfies the × overdetermined equations ∂ Φ(ζ;σ)= AΦ(ζ;σ), (1.5a) ∂ζ ∂ Φ(ζ;σ)= BΦ(ζ;σ), (1.5b) ∂σ where 4iζ2 i(σ+2q2) 4ζq+2ir iζ q A = − − , B = − , (1.6) 4ζq 2ir 4iζ2+i(σ+2q2) q iζ (cid:18) − (cid:19) (cid:18) (cid:19) and q and r are parameters which may depend on σ. It is an amiable exercise to show that the compatibility of the two equations in (1.5) is reduced to the fact that q q(σ) solves the Painlev´e ≡ equation (1.1), and the parameter r in (1.6) is r r(σ) = q (σ). ′ ≡ It is known that all solutions to the (1.1) are meromorphic, so if we choose q q(σ) to be any ≡ particular solution to (1.1) and take r q (σ), then the system (1.5) is solvable provided σ is not ′ ≡ a pole of the chosen PII transcendent. Notice then that, given a particular solution q(σ) and fixing σ that is not a pole of this solution, we can find a solution to the overdetermined equation (1.5) using only (1.5a), given proper initial conditions. Thus below we concentrate on (1.5a) when we talk about the solutions to (1.5), where q(σ) is a fixed solution to (1.1), r(σ) = q (σ), and σ is a ′ constant that is not a pole of q. In some formulas in this paper, we suppress the dependence on σ if it is treated as a constant. Since is the only singular point of A, and ∞ 4iζ2 0 A= (I +O(ζ−1)) −0 4iζ2 , as ζ → ∞, (1.7) (cid:18) (cid:19) it is natural to construct the fundamental solution Φ such that Φ(ζ)= (I + (ζ 1)) e−43iζ3−iσζ 0 , as ζ . (1.8) O − 0 e43iζ3+iσζ! → ∞ But is an irregular singularity of A, so the Stokes phenomenon allows us only to consider the ∞ solution Φ that satisfies (1.8) sectorally. For a rigorous version of the heuristic argument above see [17, Section 5.0]. For j = 0,1,...,5, define the sectors (see Figure 1), π jπ π jπ S = z C: + < argz < + . (1.9) j ∈ −6 3 6 3 (cid:26) (cid:27) Their boundaries are the rays with outward orientation Σk = te(k−1/2)i3π t [0, ) , k = 0,...,5. (1.10) | ∈ ∞ n o 3 Then there are fundamental solutions Ψ(0),...,Ψ(5) to (1.5a) such that Ψ(j) satisfies the boundary condition (1.8) in sector S . Of course the solution space to (1.5a) is two dimensional and so there j are linear relations between the solutions Ψ(0),...,Ψ(5). These relations depend on the particular Painlev´e transcendent appearing in the coefficient matrices A and B in (1.6), and can be described in the following way [17, Section 5.0]. For each PII solution q(σ) to (1.1), there is a triple of complex numbers (t ,t ,t ) satisfying 1 2 3 the relation t +t +t +t t t = 0, (1.11) 1 2 3 1 2 3 such that the fundamental solutions Ψ(k) associated with q(σ) satisfy Ψ(k) = Ψ(k 1)J , k = 0,...,5, with J shown in Figure 1 and Ψ( 1) := Ψ(5). (1.12) − k k − The jump matrices J are called the Stokes matrices, and the numbers t ,t ,t are called the k 1 2 3 Stokes multipliers corresponding to the given PII solution q(σ). Remarkably, each triple (t ,t ,t ) 1 2 3 of Stokes multipliers satisfying (1.11) corresponds uniquely to a PII solution, and so the solutions to PII are parametrized by the surface (1.11). Thus in order to specify a solution to PII, it is enough to specify the Stokes multipliers (t ,t ,t ), see [17, Proposition 5.1]. In Figure 1 we show 1 2 3 the rays, sectors, and the jump matrices J . k J2=(cid:18)10 t12(cid:19) Ψ(1)= J3= S2 S1 J1= Ψ(2)=(cid:0)ψ(1)+t1ψ(2),t2ψ(1)+(t1t2+1)ψ(2)(cid:1) ψ(1)+t1ψ(2),ψ(2) 1 0 Σ2 1 0 (cid:16) (cid:17) (cid:18)t3 1(cid:19) Σ1 (cid:18)t1 1(cid:19) Ψ(3)= S3 Σ3 S0 (cid:16)(t2t3+1)ψ(1)−t2ψ(2),t2ψ(1)+(t1t2+1)ψ(2)(cid:17) Ψ(0)= Σ0 ψ(1),ψ(2) (cid:16) (cid:17) J4= Σ4 J0= (cid:18)10 t11(cid:19) S4 Σ5 S5 (cid:18)10 t13(cid:19) Ψ(5)= J5=(cid:18)t12 01(cid:19) Ψ(4)=(cid:0)(t2t3+1)ψ(1)−t2ψ(2),−t3ψ(1)+ψ(2)(cid:1) (cid:16)ψ(1),−t3ψ(1)+ψ(2)(cid:17) Figure 1: Rays Σ , sectors S , and Figure2: Theformulas of Ψ(0),...,Ψ(5) expressedinψ(1) and k k jump matrices J placed on Σ for ψ(2). k k k = 0,...,5. For a given set of Stokes multipliers, the jump properties (1.12) determine any of the funda- mental solutions in terms of the solution Ψ(0). Indeed if we denote Ψ(0)(ζ;σ) = ψ(1)(ζ;σ),ψ(2)(ζ;σ) , (1.13) (cid:16) (cid:17) where ψ(1) and ψ(2) are two 2-dimensional vector-valued functions defined on the whole complex plane, then the other Ψ(k) are expressed in ψ(1) and ψ(2) as in Figure 2. The asymptotics of the 4 columns of Ψ(k) are summarized below (with δ being any small positive constant): ψ(1)(ζ) if arg(ζ) ( 2π +δ, π δ), ψ(1)(ζ)+t1ψ(2)(ζ) = (I + (ζ−1)) e−43iζ3−iσζ if arg(ζ)∈ (−δ,π3 δ),3 −  O 0 ∈ − (t t +1)ψ(1)(ζ) t ψ(2)(ζ) (cid:18) (cid:19) if arg(ζ) (2π +δ, 5π δ), 2 3 − 2 ∈ 3 3 −   (1.14)    ψ(2)(ζ) if arg(ζ) ( π +δ, 2π δ), 0 ∈ −3 3 − −t3ψ(1) +ψ(2)(ζ) = (I +O(ζ−1)) e43iζ3+iσζ if arg(ζ)∈ (π+δ,2π −δ), t ψ(1) +(t t +1)ψ(2)(ζ) (cid:18) (cid:19) if arg(ζ) (π +δ, 4π δ). 2 1 2 ∈ 3 3 −   (1.15)    1.1.1 Critical kernel in one-matrix model As mentioned earlier, the Hastings–McLeod solution to (1.1), the one satisfying (1.2), is of special importance in random matrix theory. It is the solution to PII that corresponds to the Stokes multipliers (t ,t ,t ) = (1,0, 1), and it is well established that it has no poles on the real line. 1 2 3 − Thus the solution Ψ(0) Ψ(0)(ζ;σ) exists for any real σ [17, Section 11.7]. ≡ Considertheone-matrixmodelgivenbytheprobabilitymeasureonthespaceofn nHermitian × matrices M, 1 exp( ntTrV(M))dM, (1.16) C − n where V is the potential and t > 0 is a scaling factor. The eigenvalues of M are a determinantal processthatis characterized byacorrelation kernel. InthecasethatV(x) = x4/4 x2 andn , − → ∞ themodelisinacriticalphaseift = 1. Asn ,underthedoublescalinglimitt = 1 (2n) 2/3σ, − → ∞ − the correlation kernel at u(n/4) 1/3 and v(n/4) 1/3 converges to − − (1) (1) (1) (1) ψ (u;σ)ψ (v;σ)+ψ (u;σ)ψ (v;σ) Kcr(u,v;σ) = − 1 2 2 1 , (1.17) 1 2πi(u v) − where ψ(1) and ψ(1) are the two components of the 2-vector ψ(1) defined in (1.13), see [4]. We use 1 2 the notation Kcr to emphasize that this kernel arises in a 1-matrix model and to differentiate it 1 from the kernel (1.74) which arises in a 2-matrix model, which we denote Kcr. Note that although 2 we only state the limiting correlation kernel for a very special potential function, the convergence to Kcr holds for a large class of potentials that have a quadratic interior critical point. See [6] for 1 the universality of the limiting kernel Kcr. 1 Finally we remark that if we give the potential V a logarithmic perturbation at 0, i.e., let V(x) = x4/4 x2 (2α/n)log x , then the limiting kernel at 0 is changed, and it is expressed by − − | | theFlaschka–Newell LaxpairfortheHastings–McLeodsolutionoftheinhomogeneous PIIequation. See [5] for detail, and also see Section 1.5. 1.2 A 4 4 Lax system for PII × Now we introduce the other Lax system for the PII equation (1.1), which was discovered recently by Delvaux, Kuijlaars, and Zhang in their study of non-intersecting Brownian motions [12], by 5 Delvaux in the study of non-intersecting squared Bessel processes [9], and by Duits and Geudens in their study of the 2-matrix model [14], see also [10], [26]. In its most general form this Lax system is a 4-dimensional overdetermined differential system consisting of 16 equations. Here we consider a 4 4 matrix valued function M = M(z,s ,s ,τ), and the Lax system is 1 2 × ∂ M = UM, (1.18a) ∂z ∂ ∂ ∂ M =V M, M = V M, M = WM. (1.18b) 1 2 ∂s ∂s ∂τ 1 2 The coefficient matrix U is given by U11 U12 U = , (1.19) U21 U22 (cid:18) (cid:19) where each Uij is a 2 2 block, such that × U11 = τ −s21+ Cu γ√Cr√2rq1 , U12 = ir1 0 , U22 = τ +s21− Cu γ√√rr12qC , γ√r1q τ +s2 u! 0 ir2 γ √r2q τ s2+ u! − C√r2 − 2− C (cid:18) (cid:19) − √r1C − − 2 C U21 = i r1z−2s1+ rs141 − 2rs121Cu + ur21−Cq22 √r1r2Cγ(q′+uq) − (γrC12s(22r+1rr222)s321/)2q . γ√r1r2C(q′+uq)− γ(Cr12(sr221+r2r)223s/212)q −r2z−2s2+ sr242 − 2rs222Cu + ur22−Cq22   (1.20) Here the numbers r and r are positive constants, and C, γ, q, q , and u depend on r ,r ,s ,s ,τ. 1 2 ′ 1 2 1 2 We relegate the formulas for V ,V ,W to Appendix A, since we do not use them in the rest of this 1 2 paper. In the symmetric case r = r and s = s , see also [9, Section 5.3], [14], and in the τ = 0 1 2 1 2 case, see also [12, Section 5.2]. By the compatibility of the overdetermined system, which is routine but laborious, see [10, Section 6.5], we derive 8 r2 r2 r s r s C = (r1−2+r2−2)1/3, γ = exp 3(r21+−r22)2τ3−4 1r12−+r22 2τ , (1.21) (cid:18) 1 2 1 2 (cid:19) and q and u are functions of 2 s s 2τ2 1 2 σ := + . (1.22) C r r − r2+r2 (cid:18) 1 2 1 2(cid:19) Furthermore, q = q(σ) satisfies the PII equation (1.1), q = q (σ) is the derivative with respect to ′ ′ σ, and u is the PII Hamiltonian u(σ) := q (σ)2 q(σ)2 q(σ)4, (1.23) ′ − − which satisfies u(σ) = q(σ)2. (1.24) ′ − Now as with the Lax pair (1.5), we fix a particular solution q(σ) to PII and assume σ is not a pole of this solution. We can then solve the Lax system by (1.18a) alone, with proper initial conditions. 6 Remark 1.3. The authors of [12], [14], [10], and [9] introduced the Lax system (1.18) as a technical tool to study the tacnode Riemann–Hilbert problem (RHP), a 4 4 Riemann–Hilbert problem × associated with the PII equation (1.1). The tacnode RHP is only defined for the Hastings–McLeod solutiontoPII,buttheLaxsystemisalgebraicandtheFrobeniuscompatibility conditions(1.4)are independent of boundary condition, so the Lax system exists for all solutions to the PII equation. From the Lax system we can construct an RHP that is associated with all solutions to the PII equation and thus generalize the tacnode RHP. See Riemann–Hilbert problem 1.5 in Section 1.4.3 below. Since is the unique singular point of U, it is natural to put the boundary condition to the ∞ solution M at . The situation is a bit more complicated than for the 2 2 Lax system, since ∞ × infinity is, in the language of [17], a general irregular singular point of the coefficient matrix U. Nonetheless, it is possible to transform the equation (1.18a) into one with a regular singular point by means of an explicit change of variable, and then to derive the asymptotic structure of its solutions using the methods of [31]. This asymptotic structure was derived by Duits and Geudens in [14]. In order to describe it, we define the functions 2 θ1(z) = r1( z)32 +2s1( z)12, z C [0, ), 3 − − ∈ \ ∞ (1.25) 2 θ2(z) = r2z32 +2s2z12, z C ( ,0], 3 ∈ \ −∞ and then the 4-dimensional vector-valued functions 1 T 1 T v1(z) = e−θ1(z)+τz ( z)−14,0, i( z)14,0 , v2(z) = e−θ2(z)−τz 0,z−14,0,iz14 , √2 − − − √2 (cid:16) (cid:17) (cid:16) (cid:17) 1 T 1 T v3(z) = eθ1(z)+τz i( z)−41,0,( z)14,0 , v4(z) = eθ2(z)−τz 0,iz−14,0,z14 , √2 − − − √2 (cid:16) (cid:17) (cid:16) (cid:17) (1.26) and the matrix-valued function (z) := v (z),v (z),v (z),v (z) . (1.27) 1 2 3 4 A (cid:16) (cid:17) For the fractional powers in (1.26) we take the principal branches, so (z) has cuts on R and + A R . More precisely, the functions v (z) and v (z) each have cuts on the positive real axis, and 1 3 − the functions v (z) and v (z) each have cuts on the negative real axis. We also define the function 2 4 +(z)tobethecontinuationof (z)fromtheupperhalfplanewithacutonthenegativeimaginary A A axis, and and (z) to be the continuation of (z) from the lower half plane with a cut on the − A A positive imaginary axis. To be concrete, we denote (z) = v (z),v (z),v (z),v (z) , (1.28) A± 1± 2± 3± 4± (cid:16) (cid:17) such that for all j = 1,...,4, v (z) = v (z) in C , and the branch cut for v (z) is it t 0 . If j± j ± j± {∓ | ≥ } wedenote by v+(z) (resp. v (z)) thelimiting valueof v (z) fromthe upper(resp. lower) half-plane j j− j for j = 1,2,3,4, then we have the following relations: v+(z) = v (z) and v+(z) = v (z), z R , 1 − 3− 3 1− ∈ + (1.29) v+(z) = v (z) and v+(z) = v (z), z R . 2 − 4− 4 2− ∈ − 7 Again due to the Stokes phenomenon, we cannot find solutions that satisfy the boundary con- ditions at from all directions, but only sectorally. Here we follow the notation in [14] and define ∞ six overlapping sectors in the complex plane π jπ 7π jπ Ω := z C : + < argz < + , j = 0,...,5, (1.30) j ∈ −12 3 12 3 (cid:26) (cid:27) as shown in Figure 5. The following result was proved in [14, Lemma 5.2]. Proposition 1.1. For fixed r ,r > 0 and Ω one of the sectors defined in (1.30), the equation 1 2 j (1.18a) has a unique fundamental solution M(j) such that as z within Ω , j → ∞ I + (z 1) +(z), for j = 0,1,2, M(j)(z) = O − A (1.31) ((cid:0)I + (z−1)(cid:1) −(z), for j = 3,4,5. O A Remark 1.4. In [14, Lemma 5.2], the(cid:0)above resul(cid:1)t is stated for s = s R, and r = r = 1, but 1 2 1 2 ∈ it is trivial to extend to the more general parameters s ,s ,τ and r ,r > 0. 1 2 1 2 Remark 1.5. The general theory outlined in [31, Theorem 19.1] would indicate a weaker result, namely an asymptotic expansion in powers of z 1/2 rather than in powers of z 1. The stronger − − asymptotics above are the result of some symmetry in the equation (1.18a), see the proof in [14]. Below we construct six 4-vector-valued functions solutions to ∂ m = Um, (1.32) ∂z which we denote by n(0),...,n(5), explicitly from the solutions to the Flaschka–Newell Lax pair (1.5a). It is then shown that the solution n(j) is recessive in the sector S which was defined in j (1.9). Thus these solutions comprise the essential components of the fundamental solutions M(j) satisfying (1.31). 1.3 Main results Inorder tostate theconstruction andpropertiesof n(0),...,n(5), wefirstintroducesome notations. Suppose Γ = Γ Γ Γ is a trivalent contour, where Γ ,Γ , and Γ are three rays in the complex 1 2 3 1 2 3 ∪ ∪ plane which meet at the origin such that Γ and Γ are oriented away from the origin, and Γ is 1 2 3 oriented towards the origin. Denote a,b, c, γ , and γ as 1 2 4 r2 r2 8τ 1 4τ2(r2 r2) s s a = 1 − 2 , b= , c= 1 − 2 2 1 2 , 3 r2+r2 C2(r2+r2) C (r2+r2)2 − r − r (cid:18) 1 2(cid:19) 1 2 (cid:20) 1 2 (cid:18) 1 2(cid:19)(cid:21) (1.33) 8r4τ3 4r s τ 8r4τ3 4r s τ γ = exp 1 + 1 1 , γ = exp 2 + 2 2 , 1 −3(r2+r2)3 r2+r2 2 −3(r2+r2)3 r2+r2 (cid:18) 1 2 1 2(cid:19) (cid:18) 1 2 1 2(cid:19) and then the function G(ζ) = exp iaζ3+bζ2+icζ , (1.34) and the related functions (cid:0) (cid:1) 2 γ 2 γ 2i 2i 1 2 G (ζ) = G(ζ), G (ζ)= G(ζ), G (ζ)= ζG (ζ), G (ζ)= ζG (ζ). 1 2 3 1 4 2 πC√r πC√r C C r 1 r 2 (1.35) 8 Define now an integral transform that transforms two 2-dimensional vector-valued functions Γ Q f(ζ)= (f (ζ),f (ζ))T and g(ζ) = (g (ζ),g (ζ))T into a 4-dimensional vector-valued function given 1 2 1 2 by, (f,g)(z) := Γ Q 2izζ 2izζ 2izζ Γ1e C f1(ζ)G1(ζ)dζ + Γ2e C g1(ζ)G1(ζ)dζ + Γ3e C (f1(ζ)+g1(ζ))G1(ζ)dζ 2izζ 2izζ 2izζ RΓ1e C f2(ζ)G2(ζ)dζ +RΓ2e C g2(ζ)G2(ζ)dζ +RΓ3e C (f2(ζ)+g2(ζ))G2(ζ)dζ, (1.36) MRΓ1e22iiCzzζζf1(ζ)G3(ζ)dζ +RΓ2e22iiCzzζζg1(ζ)G3(ζ)dζ +RΓ3e22iiCzzζζ(f1(ζ)+g1(ζ))G3(ζ)dζ RΓ1e C f2(ζ)G4(ζ)dζ +RΓ2e C g2(ζ)G4(ζ)dζ +RΓ3e C (f2(ζ)+g2(ζ))G4(ζ)dζ   where rR, r , s , s , τ, and C areRthe parameters in (1.20R) and (1.21), and 1 2 1 2 1 0 0 0 M= e−τz(cid:18)rr1122−+rr2222(cid:19)ri1 τrr1122−+rr2222 +0τ −s21+ Cu ri1γ√√1rr21qC −r01i 00 , (1.37)  (cid:16) −r2iγ√√rr21Cq (cid:17) ri2 τrr1122−+rr2222 −τ +s22− Cu 0 −r2i  (cid:16) (cid:17)  where q = q(σ) is any fixed PII solution evaluated at σ defined in (1.22), and u = u(σ) is defined in (1.23). We have the following proposition. Proposition 1.2. Fix some solution q to (1.1) and let σ be as in (1.22) such that it is not a pole of q. Let φ(ζ) and ϕ(ζ) be any two 2-vector solutions to (1.5a). Assume that for a particular choice of Γ the integral transform (φ,ϕ)(z) exists and is finite for every z C. Then (φ,ϕ)(z) solves Γ Γ Q ∈ Q the differential equation (1.32). The proof of this proposition is given in Section 2. Now we make a special choice of φ(ζ) and ϕ(ζ) in Proposition 1.2 and define the particular solutions n(0),...,n(5) of (1.32). Recall the rays Σ ,...,Σ defined in (1.10) (see also Figure 1). 0 5 We define the trivalent contours Γ(0),...,Γ(5) as the Γ in Proposition 1.2 as follows: Γ(k) = Γ(k) Γ(k) Γ(k), where Γ(k) = Σ , Γ(k) = Σ , Γ(k) = ( Σ ), (1.38) 1 ∪ 2 ∪ 3 1 1−k 2 2−k 3 − 3−k where the contours Σ are oriented towards infinity, Σ means the contour Σ oriented in the j j j − opposite direction, and Σ = Σ . For an illustration of the contours, see Figure 3. i 6 i − Then we define n(k)(z) = n(k)(z;r ,r ,s ,s ,τ) = (f(k),g(k)), k = 0,...,5, (1.39) 1 2 1 2 QΓ(k) where r ,r ,s ,s ,τ are parameters in the formula of , and f(k) and g(k) are the columns of 1 2 1 2 Γ Q fundamental solutions to (1.5a), given as (1 2j) (1 2j) Ψ − (ζ;σ) Ψ − (ζ;σ)(z) f(2j)(ζ)= 1,2 , g(2j)(ζ) = t 1,1 , (1 2j) 2+j (1 2j) Ψ − (ζ;σ)! Ψ − (ζ;σ)(z)! 2,2 2,1 for j = 0,1,2, (1.40) ( 2j) ( 2j) Ψ − (ζ;σ) Ψ − (ζ;σ)(z) f(2j+1)(ζ)= 1,1 , g(2j+1)(ζ) = t 1,1 , ( 2j) 1+j ( 2j) Ψ − (ζ;σ)! Ψ − (ζ;σ)(z)! 2,1 2,1 9 wheretheparameterσ isdeterminedbytherelation (1.22),and(t ,t ,t )aretheStokes multipliers 1 2 3 correspondingto the chosen PII solution. We use the notational conventions t = t and Ψ(6+i) = 3+i i Ψ(i), and the subscripts refer to the matrix entries. Here we note that all the f(k) and g(k) are linear combinations of ψ(1) and ψ(2), as shown in Figure 3, and by the jump condition (1.12) we see that, (2 2k) (1 2k) Ψ − (ζ;σ) Ψ − (ζ;σ) f(2j)(ζ)+g(2j)(ζ;σ) = 1,2 , f(2j+1)(ζ;σ)+g(2j+1)(ζ;σ) = 1,1 , (2 2k) (1 2k) Ψ − (ζ;σ)! Ψ − (ζ;σ)! 2,2 2,1 for j = 0,1,2. (1.41) From thedefinitionsof thefunctionsn(j)(z) andtherelation (1.11), thelinear relations between themareeasy tosee, especially inFigure3. We have, forexample, thepairof independentrelations n(5)(z) = t n(0)(z) (1+t t )n(1)(z)+t n(2)(z) n(3)(z), (1.42a) 3 2 3 2 − − − n(0)(z) = t n(1)(z) (1+t t )n(2)(z)+t n(3)(z) n(4)(z). (1.42b) 2 1 2 1 − − − The next result of the paper is that the solutions n(0),...,n(5) of (1.32) satisfy the asymptotics of the columns of the fundamental solutions M(0),...,M(5) in some of the sectors Ω ,...,Ω , 0 5 and thus these fundamental solutions can be built from the columns n(0),...,n(5). To state the proposition, we recall the functions θ (z) and θ (z) defined in (1.25). 1 2 Proposition 1.3. Suppose δ > 0 is a small constant. For z = reiθ with θ fixed and r + , we → ∞ have the following asymptotic results, where all power functions take the principal branch ( π,π): − 1. Uniformly for θ ( π/3+δ,π/3 δ) ∈ − − 1 T n(0)(z) = e−θ2(z)−τz (z−43),z−14 + (z−34), (z−14),iz14 + (z−41) , (1.43) √2 O O O O (cid:16) (cid:17) uniformly for θ [0,π/3 δ) ∈ − 1 T n(2)(z) = eθ2(z)−τz (z−43),iz−14 + (z−34), (z−14),z14 + (z−14) , (1.44) −√2 O O O O (cid:16) (cid:17) and uniformly for θ ( π/3+δ,0] ∈ − 1 T n(4)(z) = eθ2(z)−τz (z−34),iz−14 + (z−34), (z−14),z14 + (z−14) . (1.45) √2 O O O O (cid:16) (cid:17) 2. Uniformly for θ (π/3+δ,π δ) ∈ − 1 T n(2)(z) = eθ2(z)−τz (z−43),iz−14 + (z−34), (z−14),z14 + (z−14) , (1.46) −√2 O O O O (cid:16) (cid:17) uniformly for θ [2π/3,π δ) ∈ − 1 T n(4)(z) = e−θ2(z)−τz (z−34),z−14 + (z−43), (z−14),iz14 + (z−14) , (1.47) −√2 O O O O (cid:16) (cid:17) and uniformly for θ (π/3+δ,2π/3] ∈ 1 T n(0)(z) = e−θ2(z)−τz (z−43),z−14 + (z−34), (z−14),iz14 + (z−41) . (1.48) √2 O O O O (cid:16) (cid:17) 10