Introduction
Because cosmological particle production in time-dependent backgrounds is numerically expensive, almost any novel computational technique is of interest. There has been some recent interest in studying particle production using Stokes phenomena techniques [1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 ] that stems from propagation in an analytically continued time coordinate in the functional space of any given choice of asymptotic expansion basis. A key ingredient in this technique stems from the fact that
h ( R e i θ ) e − R ∼ 0 (1) h(Re^{i\theta})e^{-R} \sim0 \tag*{(1)} h ( R e i θ ) e − R ∼ 0 (1)
asymptotically as R → ∞ R \to\infty R → ∞ to all orders in R − 1 R^{-1} R − 1 for almost any function h ( z ) h(z) h ( z ) that is finite in that limit, which allows h ( R e i θ ) h(Re^{i\theta}) h ( R e i θ ) to be discontinuous as a function of θ \theta θ without changing the smoothness of the asymptotically expanded function [10 , 11 , 12 , 13 , 14 ] . At the end of the computation, multiplication of judiciously weighted several of these discontinuous jumps of h ( z ) h(z) h ( z ) will become the particle production amplitude.
Many of the previous studies using the Stokes phenomena for particle production focused on the large momenta region with a variety of spins, which yields answers that are qualitatively similar to those derived from saddle-point approximations [2 , 4 , 15 ] . Reference [9 ] considered the opposite limit in which C → 0 C \to0 C → 0 in the dispersion relationship ω k 2 ( η ) ≈ C ( k ) + A η n \omega_k^2(\eta) \approx C(k) + A\eta^n ω k 2 ( η ) ≈ C ( k ) + A η n (where η \eta η is the conformal time and n n n is even) and showed that the probability amplitude β \beta β governing particle production was topological
β C → 0 + = − i cot [ π N S t o k e s s e c t o r s ] (2) \beta_{\sqrt{C}\to0+} = -i\cot\left[\frac{\pi}{N_{\mathrm{Stokes\ sectors}}}\right] \tag*{(2)} β C → 0 + = − i cot [ N Stokes sectors π ] (2)
(where N S t o k e s s e c t o r s = n + 2 N_{\mathrm{Stokes\ sectors}} = n + 2 N Stokes sectors = n + 2 ) which for example in conformally coupled scenarios corresponds to the small momenta limit. As this amplitude showcases, one of the advantages of the Stokes phenomena techniques compared to more traditional saddle-point approximations is that it allows one to compute very large coefficients [9 ] . In this paper, we will keep the leading C C C -dependent term in the expansion of β C \beta_{\sqrt{C}} β C as C → 0 C \to0 C → 0 about this topological limit to obtain analytic results for the C C C coefficient. The answer turns out to be nonperturbative in A A A and will be given for any n ∈ n \in n ∈ even in terms of an analytic formula summarized in eq. (185) .1 We use the leading C C C dependence of β C \beta_{\sqrt{C}} β C in conjunction with numerical examples to obtain a semi-analytic estimate for the number density of particles produced in eq. (179) .
The key reasoning employed to obtain our result using the technique of exponentially improved asymptotic expansion can be summarized as follows. One can obtain β C \beta_{\sqrt{C}} β C by considering θ \theta θ propagation in analytically continued time z ∼ R → ∞ R e i θ z \sim_{R\to\infty} R e^{i\theta} z ∼ R → ∞ R e i θ for certain discrete sectors of θ \theta θ . The jumps described in eq. (1) can be thought of as introducing step functions: i.e. non-analyticity in a coefficient multiplying a “zero”. This step function motivates a particular ansatz for the propagator across a single sector involving a power series of 1 / z 1/z 1/ z , exponential functions of z z z , and a θ \theta θ dependent step function. This turns the propagator solution for one of these discrete sectors in θ \theta θ into solving a recursion relationship. Expanding in small C \sqrt{C} C allows one to find a solution to the recursion relationship explicitly. This propagation solution across a single sector can be used to generate β C \beta_{\sqrt{C}} β C by employing a novel symmetry that analytically continues C \sqrt{C} C in addition to η \eta η into the complex plane.
Mathematical ideas similar to this can be found in [16 , 17 , 18 ] , but the current work provides a novel perspective and a generalization that is useful for the asymptotic basis in the present particle production physics context. Furthermore, we have expressed the automorphism structure in terms of operator maps that can be seen as a simplification of the original recursion relation presentation given by [18 ] . In the process of computing the C C C coefficient in the small C C C expansion of β C \beta_{\sqrt{C}} β C , we show that despite the asymptotic basis having non-analyticities in C ∼ k \sqrt{C} \sim k C ∼ k coming from the Lorentz group representational constraint ω = C + A η n \omega= \sqrt{C + A\eta^n} ω = C + A η n and despite this non-analyticity manifesting as a type of Schroedinger phase in intermediate expressions for β C \beta_{\sqrt{C}} β C , it disappears consistently with what one would expect from time-reversal symmetry in the system.
The order of presentation is as follows. In § 2 , we define the asymptotic basis and the analytic continuation for the particle production computation. In § 3 , we discuss the discrete symmetries (one involving complexification of both η \eta η and C \sqrt{C} C ) of the system that will help with obtaining the particle production amplitude. In § 4 , we compute the first Stokes constant for n ≥ 4 n \ge4 n ≥ 4 . The key to the computation that will be explained in this section is the exponentially improved asymptotic expansions from the perspective of an automorphism on the functional space induced by WKB amplitudes in the asymptotic region. In section 5, we combine the discrete symmetry representation and the first Stokes constant computed in § 4 to derive the coefficient c 1 c_1 c 1 in β C ≈ β 0 + c 1 C / A 2 n + 2 \beta_{\sqrt{C}} \approx\beta_0 + c_1 C/A^{\frac{2}{n+2}} β C ≈ β 0 + c 1 C / A n + 2 2 which is the main aim of this paper. There, we also use this to give a semi-analytic formula for the total number density of particles produced. We will then conclude with a brief summary. The Appendices contain many interesting technical details related to the surprisingly intricate computation of this paper.
Particle production
The objective of this section is to first set up a generic cosmological particle production scenario and review the Bogoliubov computational approach [7 , 9 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 ] . We then specialize to particle mode equations
[ ∂ η 2 + ω k 2 ( η ) ] χ k ( η ) = 0 (3) \left[\partial_{\eta}^{2}+\omega_{k}^{2}(\eta)\right]\chi_{k}(\eta)=0 \tag*{(3)} [ ∂ η 2 + ω k 2 ( η ) ] χ k ( η ) = 0 (3)
with dispersion relations of the form
ω k 2 ( η ) = C ( k ) + A η n (4) \omega_k^2(\eta) = C(k) + A\eta^n \tag*{(4)} ω k 2 ( η ) = C ( k ) + A η n (4)
where, as an extension to [9 ] , we will be interested in understanding the produced number density in the soft regime (C ( k ) / A ≪ 1 C(k)/A \ll1 C ( k ) / A ≪ 1 ). Here η \eta η is the conformal time in the Friedmann–Lemaître–Robertson–Walker (FLRW) spacetime, n n n is a positive even integer, and C ≡ C ( k ) C \equiv C(k) C ≡ C ( k ) and A A A are some real constants. Such dispersion relations also arise in many branches of physics (see e.g. [1 , 2 , 4 , 8 , 12 , 13 , 15 , 17 , 21 , 29 , 30 , 31 , 32 ] ). In § 2.1 , we briefly review particle production on time-dependent backgrounds and the relation of the produced number density to Bogoliubov coefficients. Following this, § 2.2 reviews Stokes phenomena and translates the problem of computing Bogoliubov coefficients (defined for real η \eta η ) to computing Stokes constants which characterize the analytic continuation of the solutions to eq. (3) in the complexified (asymptotic) η \eta η plane.
Setup and Bogoliubov coefficients
This subsection provides a review of the well-known Bogoliubov coefficients computation in cosmological particle production.
Consider a non-minimally coupled scalar field χ \chi χ on flat FLRW spacetime
S χ = ∫ d η d 3 x − g 2 [ ∂ μ χ ∂ μ χ − m χ 2 ( η ) χ 2 + ξ R χ 2 ] (5) S_\chi= \int d\eta d^3x \frac{\sqrt{-g}}{2} \left[\partial_\mu\chi\partial^\mu\chi- m_\chi^2(\eta)\chi^2 + \xi R\chi^2\right] \tag*{(5)} S χ = ∫ d η d 3 x 2 − g [ ∂ μ χ ∂ μ χ − m χ 2 ( η ) χ 2 + ξ R χ 2 ] (5)
d s 2 = a 2 ( η ) ( d η 2 − ∣ d x ⃗ ∣ 2 ) (6) ds^2 = a^2(\eta)\left(d\eta^2 - |d\vec{x}|^2\right) \tag*{(6)} d s 2 = a 2 ( η ) ( d η 2 − ∣ d x ∣ 2 ) (6)
where η \eta η is the conformal time and m χ ( η ) m_\chi(\eta) m χ ( η ) denotes a time-dependent effective mass due to couplings to a background field. After quantizing χ \chi χ in the FLRW spacetime [19 ] , we obtain a mode expansion
χ ( η , x ⃗ ) = ∫ d 3 k ( 2 π ) 3 a ( η ) ( a k ⃗ χ k ( η ) e i k ⃗ ⋅ x ⃗ + a k ⃗ † χ k ∗ ( η ) e − i k ⃗ ⋅ x ⃗ ) (7) \chi(\eta,\vec{x}) = \int\frac{d^3k}{(2\pi)^3a(\eta)}\left(a_{\vec{k}}\chi_k(\eta)e^{i\vec{k}\cdot\vec{x}} + a^\dagger_{\vec{k}}\chi_k^*(\eta)e^{-i\vec{k}\cdot\vec{x}}\right) \tag*{(7)} χ ( η , x ) = ∫ ( 2 π ) 3 a ( η ) d 3 k ( a k χ k ( η ) e i k ⋅ x + a k † χ k ∗ ( η ) e − i k ⋅ x ) (7)
where a k ⃗ a_{\vec{k}} a k and a k ⃗ † a^\dagger_{\vec{k}} a k † satisfy the usual commutation relations and the Fourier modes χ k ( η ) \chi_k(\eta) χ k ( η ) solve the Heisenberg equation of motion
[ ∂ η 2 + ω k 2 ( η ) ] χ k ( η ) = 0 (8) \left[\partial_\eta^2 + \omega_k^2(\eta)\right]\chi_k(\eta) = 0 \tag*{(8)} [ ∂ η 2 + ω k 2 ( η ) ] χ k ( η ) = 0 (8)
where
ω k 2 ( η ) = k 2 + a 2 m χ 2 ( η ) + ( 6 ξ − 1 ) a ′ ′ a . (9) \omega_k^2(\eta) = k^2 + a^2m_\chi^2(\eta) + (6\xi- 1)\frac{a''}{a}. \tag*{(9)} ω k 2 ( η ) = k 2 + a 2 m χ 2 ( η ) + ( 6 ξ − 1 ) a a ′′ . (9)
Here primes denote derivatives with respect to the conformal time, and the Fourier modes χ k ( η ) \chi_k(\eta) χ k ( η ) satisfy the Wronskian condition imposed by canonical quantization
− i W ( χ k , χ k ∗ ) ≡ − i ( χ k ∂ η χ k ∗ − χ k ∗ ∂ η χ k ) = 1. (10) -iW(\chi_k,\chi_k^*) \equiv-i\left(\chi_k\partial_\eta\chi_k^* - \chi_k^*\partial_\eta\chi_k\right) = 1. \tag*{(10)} − iW ( χ k , χ k ∗ ) ≡ − i ( χ k ∂ η χ k ∗ − χ k ∗ ∂ η χ k ) = 1. (10)
The vacuum state is then defined as the one annihilated by the annihilation operators a k ⃗ a_{\vec{k}} a k . Consequently, different choices of the mode functions χ k ( η ) \chi_k(\eta) χ k ( η ) corresponding to different boundary conditions (BC) imply different notions of the vacuum. A natural choice is mode functions which approximate to the positive-frequency WKB modes in time intervals where ω k ( η ) \omega_k(\eta) ω k ( η ) is adiabatic — that is, evolves slowly in η \eta η . Then, as the WKB modes limit to plane wave solutions when ω k ( η ) \omega_k(\eta) ω k ( η ) asymptotes to a constant, we may associate an approximate notion of a single particle state with the corresponding creation and annihilation operators. The corresponding vacuum is referred to as the adiabatic vacuum [19 ] .
A convenient approximate measure of nonadiabaticity (or time-translation symmetry breaking) in a time region for our dispersion relationship is
δ k ( η ) ≡ ω k ′ ( η ) 4 ω k 2 ( η ) (11) \delta_k(\eta) \equiv\frac{\omega'_k(\eta)}{4\omega_k^2(\eta)} \tag*{(11)} δ k ( η ) ≡ 4 ω k 2 ( η ) ω k ′ ( η ) (11)
corresponding to a linear dimensionless time derivative. The leading approximate WKB mode functions are given by
χ k W K B ( η ) ≈ e − i ∫ η ∗ η d η ˉ ω k ( η ˉ ) 2 ω k ( η ) (12) \chi_k^{\mathrm{WKB}}(\eta) \approx\frac{e^{-i\int_{\eta_*}^{\eta}d\bar{\eta}\omega_k(\bar{\eta})}}{\sqrt{2\omega_k(\eta)}} \tag*{(12)} χ k WKB ( η ) ≈ 2 ω k ( η ) e − i ∫ η ∗ η d η ˉ ω k ( η ˉ ) (12)
whose error is of order δ k 2 \delta_k^2 δ k 2 . For dispersion relations such as eq. (4) where the non-adiabaticities in ω k \omega_k ω k are approximately localized within a finite time interval, vacua in the asymptotic past and future of this interval may then be defined with respect to mode functions which asymptote to eq. (12) in these two regions [19 , 33 ] .
Let S \mathcal{S} S denote the space of solutions to the mode equation (8) . For the above reason, we introduce the two solutions χ k , 1 ( η ) \chi_{k,1}(\eta) χ k , 1 ( η ) , χ k , 2 ( η ) ∈ S \chi_{k,2}(\eta) \in\mathcal{S} χ k , 2 ( η ) ∈ S satisfying the WKB BC in the asymptotic past and the asymptotic future, respectively, as
lim η → η − ∞ χ k , 1 ( η ) ∼ e − i ∫ η ∗ η d η ˉ ω k ( η ˉ ) 2 ω k ( η ) , lim η → η + ∞ χ k , 2 ( η ) ∼ e − i ∫ η ∗ η d η ˉ ω k ( η ˉ ) 2 ω k ( η ) , (13) \lim_{\eta\to\eta_{-\infty}}\chi_{k,1}(\eta) \sim\frac{e^{-i\int_{\eta_*}^{\eta}d\bar{\eta}\omega_k(\bar{\eta})}}{\sqrt{2\omega_k(\eta)}}, \qquad\lim_{\eta\to\eta_{+\infty}}\chi_{k,2}(\eta) \sim\frac{e^{-i\int_{\eta_*}^{\eta}d\bar{\eta}\omega_k(\bar{\eta})}}{\sqrt{2\omega_k(\eta)}}, \tag*{(13)} η → η − ∞ lim χ k , 1 ( η ) ∼ 2 ω k ( η ) e − i ∫ η ∗ η d η ˉ ω k ( η ˉ ) , η → η + ∞ lim χ k , 2 ( η ) ∼ 2 ω k ( η ) e − i ∫ η ∗ η d η ˉ ω k ( η ˉ ) , (13)
where η − ∞ \eta_{-\infty} η − ∞ and η + ∞ \eta_{+\infty} η + ∞ denote times in the asymptotic past and future respectively, and η ∗ \eta_* η ∗ is a common reference point for the WKB phase.2 The non-adiabaticities in ω k \omega_k ω k imply that χ k , 1 \chi_{k,1} χ k , 1 and χ k , 2 \chi_{k,2} χ k , 2 are inequivalent functions such that { χ k , 1 ( η ) , χ k , 1 ∗ ( η ) } \{\chi_{k,1}(\eta),\chi^*_{k,1}(\eta)\} { χ k , 1 ( η ) , χ k , 1 ∗ ( η )} and { χ k , 2 ( η ) , χ k , 2 ∗ ( η ) } \{\chi_{k,2}(\eta),\chi^*_{k,2}(\eta)\} { χ k , 2 ( η ) , χ k , 2 ∗ ( η )} form two bases of S \mathcal{S} S to quantize χ \chi χ with. Basis independence of quantization implies these are related to each other by a linear superposition
χ k , 1 ( η ) = α k χ k , 2 ( η ) + β k χ k , 2 ∗ ( η ) , (14) \chi_{k,1}(\eta) = \alpha_k\chi_{k,2}(\eta) + \beta_k\chi^*_{k,2}(\eta), \tag*{(14)} χ k , 1 ( η ) = α k χ k , 2 ( η ) + β k χ k , 2 ∗ ( η ) , (14)
where α k \alpha_k α k and β k \beta_k β k are constants, referred to as the Bogoliubov coefficients.
What the above superposition implies in the context of the class of dispersion relations given by eq. (4) is the following. The time dependence of ω k \omega_k ω k implies that the adiabatic condition ∣ δ k ∣ ≪ 1 |\delta_k| \ll1 ∣ δ k ∣ ≪ 1 breaks down in the region ∣ η ∣ ≲ ( C / A ) 1 / n |\eta| \lesssim(C/A)^{1/n} ∣ η ∣ ≲ ( C / A ) 1/ n . Consequently, although the solution χ k , 1 ( η ) \chi_{k,1}(\eta) χ k , 1 ( η ) corresponds to a pure positive-frequency mode in the asymptotic past, the violation of adiabaticity near ∣ η ∣ ∼ ( C / A ) 1 / n |\eta| \sim(C/A)^{1/n} ∣ η ∣ ∼ ( C / A ) 1/ n causes χ k , 1 ( η ) \chi_{k,1}(\eta) χ k , 1 ( η ) to evolve into a superposition of positive and negative frequency modes in the asymptotic future. This in turn implies that the Heisenberg vacuum state ∣ p a s t ⟩ \lvert\mathrm{past}\rangle ∣ past ⟩ fixed in the asymptotic past is an excited state when measured with the Heisenberg picture Hamiltonian in the asymptotic future. The consequently produced particle number density given by
n χ ≡ ⟨ p a s t ∣ n χ ( f u t u r e ) ∣ p a s t ⟩ = 1 a 3 ∫ d 3 k ( 2 π ) 3 ∣ β k ∣ 2 (15) n_{\chi} \equiv\langle\mathrm{past}|n_{\chi}^{(\mathrm{future})}|\mathrm{past}\rangle= \frac{1}{a^3}\int\frac{\mathrm{d}^3 k}{(2\pi)^3}|\beta_k|^2 \tag*{(15)} n χ ≡ ⟨ past ∣ n χ ( future ) ∣ past ⟩ = a 3 1 ∫ ( 2 π ) 3 d 3 k ∣ β k ∣ 2 (15)
is obtained by computing the expectation value of the Heisenberg picture number density operator n χ ( f u t u r e ) n_{\chi}^{(\mathrm{future})} n χ ( future ) made of normal ordered creation/annihilation operator that annihilates the future vacuum state ∣ f u t u r e ⟩ \lvert\mathrm{future}\rangle ∣ future ⟩ (where n χ ( f u t u r e ) ∣ f u t u r e ⟩ ≈ 0 n_{\chi}^{(\mathrm{future})}\lvert\mathrm{future}\rangle\approx0 n χ ( future ) ∣ future ⟩ ≈ 0 ).
In scenarios where the relevant mode functions can be computed — either analytically or numerically — the Bogoliubov coefficient β k \beta_k β k can be computed by projecting eq. (14) onto χ k , 2 ( η ) \chi_{k,2}(\eta) χ k , 2 ( η ) using the Wronskian. Recall that the Wronskian provides the normalization (10) and vanishes for linearly dependent solutions to eq. (8) . This projection leads to
β k = i W ( χ k , 1 , χ k , 2 ) = i [ χ k , 1 ( η ) ∂ η χ k , 2 ( η ) − χ k , 2 ( η ) ∂ η χ k , 1 ( η ) ] . (16) \beta_k = iW(\chi_{k,1},\chi_{k,2}) = i[\chi_{k,1}(\eta)\partial_\eta\chi_{k,2}(\eta) - \chi_{k,2}(\eta)\partial_\eta\chi_{k,1}(\eta)]. \tag*{(16)} β k = iW ( χ k , 1 , χ k , 2 ) = i [ χ k , 1 ( η ) ∂ η χ k , 2 ( η ) − χ k , 2 ( η ) ∂ η χ k , 1 ( η )] . (16)
Because the nontrivial time dependences in the system often preclude an analytic form of the mode functions, an accurate estimate of β k \beta_k β k typically requires expensive numerical computations. Hence, analytic treatments, such as saddle-point techniques, which may be used to estimate β k \beta_k β k in certain simplified scenarios are useful. The remainder of this paper is devoted to answering one such question: for the mode equation given by eqs. (3) and (4) , how can one derive an analytic expression for soft corrections to the Bogoliubov coefficient β k \beta_k β k ? As a preliminary step, we review the formalism of the Stokes phenomena approach and establish our conventions in the next section.
Analytic continuation and Stokes phenomena
Stokes phenomenon provides an approach to compute the Bogoliubov coefficients based on the analyticity properties of the dispersion relation ω k \omega_k ω k . We noted in § 2.1 that particle production may be understood in terms of the non-adiabatic evolution of the positive frequency WKB mode (12) in the asymptotic past into a superposition of positive and negative frequency WKB modes in the asymptotic future. The superposition coefficients, given by the Bogoliubov coefficients, then determine the produced particle number density. Therefore, the problem of computing particle production is equivalent to that of tracking the evolution of the WKB modes across non-adiabaticities in ω k \omega_k ω k — similar to scattering problems in quantum mechanics. Stokes phenomenon describes exactly this evolution, but in the analytically continued version of this problem.
Continuing the conformal time η \eta η to z ∈ C z \in\mathbb{C} z ∈ C results in the complexified mode equation
[ ∂ z 2 + ω k 2 ( z ) ] χ k ( z ) = 0 (17) [\partial_z^2+\omega_k^2(z)]\chi_k(z)=0 \tag*{(17)} [ ∂ z 2 + ω k 2 ( z )] χ k ( z ) = 0 (17)
where ω k ( z ) \omega_k(z) ω k ( z ) and χ k ( z ) \chi_k(z) χ k ( z ) are the analytic continuations of their real time versions. Similarly, we may define the complexified WKB basis functions as
f ± ( k , z ) ≡ e ± i θ ( k , z ) 2 ω k ( z ) , θ ( k , z ) ≡ ∫ η ∗ z d z ′ ω k ( z ′ ) (18) f_{\pm}(k,z) \equiv\frac{e^{\pm i\theta(k,z)}}{\sqrt{2\omega_k(z)}}, \qquad\theta(k,z) \equiv\int_{\eta_*}^{z}\mathrm{d}z'\,\omega_k(z') \tag*{(18)} f ± ( k , z ) ≡ 2 ω k ( z ) e ± i θ ( k , z ) , θ ( k , z ) ≡ ∫ η ∗ z d z ′ ω k ( z ′ ) (18)
where θ ( k , z ) \theta(k,z) θ ( k , z ) is the WKB phase.3 For the WKB approximation4 of a given mode function χ k ( z ) \chi_k(z) χ k ( z ) in the ‘adiabatic’ regions of the complex plane defined by
∣ δ k ( z ) ∣ = ∣ ω k ′ ( z ) 4 ω k 2 ( z ) ∣ ≪ 1 , (19) \lvert\delta_k(z)\rvert=\left\lvert\frac{\omega'_k(z)}{4\omega_k^2(z)}\right\rvert\ll1, \tag*{(19)} ∣ δ k ( z )∣ = 4 ω k 2 ( z ) ω k ′ ( z ) ≪ 1 , (19)
Stokes phenomenon implies that the superposition coefficients (defined explicitly in eq. (32) ) will shift discretely across certain contours in the complex plane called the Stokes lines. These are contours along which the WKB phase is such that i θ ( k , z ) i\theta(k,z) i θ ( k , z ) is real. Hence, along these, the two WKB basis functions f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) which either diverge or converge exponentially as ∣ z ∣ \lvert z\rvert ∣ z ∣ increases, are therefore maximally dominant or subdominant with respect to each other. These contours are further classified as “+”(“−”) Stokes lines if f + ( k , z ) f_+(k,z) f + ( k , z ) (f − ( k , z ) f_-(k,z) f − ( k , z ) ) is the dominant WKB function.
Given this, the statement of Stokes phenomenon may be understood as a consequence of the mismatch between the basis functions f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) and the exact mode χ k ( z ) \chi_k(z) χ k ( z ) as well as the peculiar property of the zero asymptotic expansion representation of exponentially suppressed functions as ∣ z ∣ → ∞ \lvert z\rvert\to\infty ∣ z ∣ → ∞ . In adiabatic regions where f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) present good approximate solutions to eq. (17) , this mismatch between a given χ k ( z ) \chi_k(z) χ k ( z ) and f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) arises as shifts in the WKB superposition coefficients as χ k ( z ) \chi_k(z) χ k ( z ) is analytically continued across the Stokes lines. These shifts occur only in the coefficient of the exponentially suppressed WKB function which in the limit ∣ z ∣ → ∞ \lvert z\rvert\to\infty ∣ z ∣ → ∞ , ∣ ω k ∣ → ∞ \lvert\omega_k\rvert\to\infty ∣ ω k ∣ → ∞ makes zero contribution to the WKB approximation truncated at O ( z − 1 ) O(z^{-1}) O ( z − 1 ) . This preserves the continuity of χ k ( z ) \chi_k(z) χ k ( z ) in its WKB approximation.
The constants which determine the above shifts are called Stokes constants. In the following, we determine the relation between the Bogoliubov coefficients and the relevant Stokes constants in the context of dispersion relations of the form eq. (4) , where we have set C = k 2 C=k^2 C = k 2 for physical clarity. This replaces the problem of computing β k \beta_k β k with that of computing the Stokes constants which we will turn to in sections 3 3 3 and 4 4 4 .
To formalize the idea of tracking the evolution of the analytically continued WKB modes, we begin by first rewriting χ k ( η ) \chi_k(\eta) χ k ( η ) on the real line as
χ k ( η ) = α k ( η ) f − p h y ( k , η ) + β k ( η ) f + p h y ( k , η ) , η ∈ R (20) \chi_k(\eta)=\alpha_k(\eta)f^{\mathrm{phy}}_-(k,\eta)+\beta_k(\eta)f^{\mathrm{phy}}_+(k,\eta),\qquad\eta\in\mathbb{R} \tag*{(20)} χ k ( η ) = α k ( η ) f − phy ( k , η ) + β k ( η ) f + phy ( k , η ) , η ∈ R (20)
where f ± p h y ( k , η ) f^{\mathrm{phy}}_\pm(k,\eta) f ± phy ( k , η ) are the physical WKB basis functions defined as
f ± p h y ( k , η ) ≡ e ± i θ k ( η ) 2 ω k ( η ) , θ k ( η ) ≡ ∫ 0 η d η ′ ω k ( η ′ ) (21) f^{\mathrm{phy}}_\pm(k,\eta)\equiv\frac{e^{\pm i\theta_k(\eta)}}{\sqrt{2\omega_k(\eta)}},\qquad\theta_k(\eta)\equiv\int_0^\eta\mathrm{d}\eta'\,\omega_k(\eta') \tag*{(21)} f ± phy ( k , η ) ≡ 2 ω k ( η ) e ± i θ k ( η ) , θ k ( η ) ≡ ∫ 0 η d η ′ ω k ( η ′ ) (21)
and the functions α k ( η ) \alpha_k(\eta) α k ( η ) and β k ( η ) \beta_k(\eta) β k ( η ) will be called the WKB amplitudes. Here both ω k ( η ) \omega_k(\eta) ω k ( η ) and ω k ( η ) \sqrt{\omega_k(\eta)} ω k ( η ) are non-negative for all η ∈ R \eta\in\mathbb{R} η ∈ R as remarked in footnote 2. Consequently, the physical WKB basis functions asymptote to the WKB BCs defined in eq. (13) , therefore the Bogoliubov coefficients from eq. (14) are related to the WKB amplitudes α k ( η ) \alpha_k(\eta) α k ( η ) and β k ( η ) \beta_k(\eta) β k ( η ) via
( α k , β k ) = lim η → η + ∞ ( α k ( η ) , β k ( η ) ) . (22) (\alpha_k,\beta_k)=\lim_{\eta\to\eta_{+\infty}}(\alpha_k(\eta),\beta_k(\eta)). \tag*{(22)} ( α k , β k ) = η → η + ∞ lim ( α k ( η ) , β k ( η )) . (22)
Henceforth, we set the common reference point η ∗ \eta_* η ∗ of the WKB phase (see eq. (18) ) to 0 for the ease of describing symmetry representations in § 3 .
Note that in rewriting the mode function χ k ( η ) \chi_k(\eta) χ k ( η ) in terms of the WKB amplitudes we have expressed one complex degree of freedom in terms of two, allowing multiple functions α k ( η ) \alpha_k(\eta) α k ( η ) and β k ( η ) \beta_k(\eta) β k ( η ) to represent the same χ k ( η ) \chi_k(\eta) χ k ( η ) . This additional freedom can be fixed by specifying χ k ′ ( η ) \chi_k'(\eta) χ k ′ ( η ) in terms of the WKB functions and amplitudes. A convenient choice is
∂ η χ k ( η ) = i ω k ( η ) ( β k ( η ) f + p h y ( k , η ) − α k ( η ) f − p h y ( k , η ) ) (23) \partial_\eta\chi_k(\eta)=i\omega_k(\eta)\left(\beta_k(\eta)f_+^{\mathrm{phy}}(k,\eta)-\alpha_k(\eta)f_-^{\mathrm{phy}}(k,\eta)\right) \tag*{(23)} ∂ η χ k ( η ) = i ω k ( η ) ( β k ( η ) f + phy ( k , η ) − α k ( η ) f − phy ( k , η ) ) (23)
which allows the second order ODE (17) of χ k ( η ) \chi_k(\eta) χ k ( η ) to be rewritten as the first order vector ODE of the WKB amplitudes (see eq. (37) in [9 ] )
∂ η V k ( η ) = M 0 ( k , η ) V k ( η ) , V k ( η ) ≡ ( α k ( η ) β k ( η ) ) , (24) \partial_\eta V_k(\eta)=M_0(k,\eta)V_k(\eta),\qquad V_k(\eta)\equiv\begin{pmatrix}\alpha_k(\eta)\\\beta_k(\eta)\end{pmatrix}, \tag*{(24)} ∂ η V k ( η ) = M 0 ( k , η ) V k ( η ) , V k ( η ) ≡ ( α k ( η ) β k ( η ) ) , (24)
M 0 ( k , η ) = ω k ′ ( η ) 2 ω k ( η ) ( 0 e + 2 i ∫ η ∗ η d η ′ ω k ( η ′ ) e − 2 i ∫ η ∗ η d η ′ ω k ( η ′ ) 0 ) . (25) M_0(k,\eta)=\frac{\omega_k'(\eta)}{2\omega_k(\eta)}
\begin{pmatrix}
0 & e^{+2i\int_{\eta_*}^{\eta}d\eta'\omega_k(\eta')}\\
e^{-2i\int_{\eta_*}^{\eta}d\eta'\omega_k(\eta')} & 0
\end{pmatrix}. \tag*{(25)} M 0 ( k , η ) = 2 ω k ( η ) ω k ′ ( η ) ( 0 e − 2 i ∫ η ∗ η d η ′ ω k ( η ′ ) e + 2 i ∫ η ∗ η d η ′ ω k ( η ′ ) 0 ) . (25)
Integrating the above we see that the propagator which relates the WKB amplitudes at two times η 0 \eta_0 η 0 and η \eta η is given by
V k ( η ) = U 0 ( k , η , η 0 ) V k ( η 0 ) (26) V_k(\eta)=U_0(k,\eta,\eta_0)V_k(\eta_0) \tag*{(26)} V k ( η ) = U 0 ( k , η , η 0 ) V k ( η 0 ) (26)
U 0 ( k , η , η 0 ) = T [ I 2 × 2 exp ( ∫ η 0 η d w M 0 ( k , w ) ) ] (27) U_0(k,\eta,\eta_0)=\mathcal{T}\left[I_{2\times2}\exp\left(\int_{\eta_0}^{\eta}dwM_0(k,w)\right)\right] \tag*{(27)} U 0 ( k , η , η 0 ) = T [ I 2 × 2 exp ( ∫ η 0 η d w M 0 ( k , w ) ) ] (27)
where T \mathcal{T} T denotes time-ordering. Note, since M 0 ( k , η ) M_0(k,\eta) M 0 ( k , η ) belongs to the Lie algebra of SU( 1 , 1 ) (1,1) ( 1 , 1 ) (see eq. (22) in [9 ] ), the WKB amplitudes on the real line satisfy
∣ α k ( η ) ∣ 2 − ∣ β k ( η ) ∣ 2 = 1 , ∀ η ∈ R . (28) |\alpha_k(\eta)|^2-|\beta_k(\eta)|^2=1,\qquad\forall\eta\in\mathbb{R}. \tag*{(28)} ∣ α k ( η ) ∣ 2 − ∣ β k ( η ) ∣ 2 = 1 , ∀ η ∈ R . (28)
This naturally extends the normalization condition on the Bogoliubov coefficients arising from eq. (10) to the WKB amplitudes at all real times.5 We will see in § 3 and § 4 that the choice of eq. (23) and eq. (25) will be convenient for analyzing discrete symmetry representations and asymptotic behaviors of the analytically continued WKB amplitudes.
Now, to extend the problem to the complex z z z -plane, we may analytically continue the first order ODE in eq. (24) and the propagator of eq. (27) . The propagators U 0 ( k , z , z 0 ) U_0(k,z,z_0) U 0 ( k , z , z 0 ) then relate the analytically continued WKB amplitudes between two complex points z z z and z 0 z_0 z 0 . Specializing to dispersion relations of the form (4) , the mode equation becomes
[ ∂ z 2 + ω k ˉ 2 ( z ) ] χ k ˉ ( z ) = 0 , ω k ˉ 2 ( z ) = k ˉ 2 + z n , z ∈ C (29) \left[\partial_z^2+\omega_{\bar{k}}^2(z)\right]\chi_{\bar{k}}(z)=0,\qquad\omega_{\bar{k}}^2(z)=\bar{k}^2+z^n,\qquad z\in\mathbb{C} \tag*{(29)} [ ∂ z 2 + ω k ˉ 2 ( z ) ] χ k ˉ ( z ) = 0 , ω k ˉ 2 ( z ) = k ˉ 2 + z n , z ∈ C (29)
where n n n is a positive even integer and the complexified conformal time has been rescaled z = A 1 / ( n + 2 ) η z = A^{1/(n+2)}\eta z = A 1/ ( n + 2 ) η to absorb the constant A A A resulting in k ˉ = C 1 / 2 / A 1 / ( n + 2 ) \bar{k} = C^{1/2}/A^{1/(n+2)} k ˉ = C 1/2 / A 1/ ( n + 2 ) .6 If we define χ ˉ k ˉ , j ( z ) \bar{\chi}_{\bar{k},j}(z) χ ˉ k ˉ , j ( z ) to be the mode equation solution satisfying eq. (13) with η → z \eta\to z η → z , one can easily check eq. (16) becomes
χ ˉ k ˉ , 1 ( z ) ∂ z χ ˉ k ˉ , 2 ( z ) − χ ˉ k ˉ , 2 ( z ) ∂ z χ ˉ k ˉ , 1 ( z ) = χ k , 1 ( η ) ∂ η χ k , 2 ( η ) − χ k , 2 ( η ) ∂ η χ k , 1 ( η ) (30) \bar{\chi}_{\bar{k},1}(z)\partial_z\bar{\chi}_{\bar{k},2}(z)-\bar{\chi}_{\bar{k},2}(z)\partial_z\bar{\chi}_{\bar{k},1}(z)=\chi_{k,1}(\eta)\partial_\eta\chi_{k,2}(\eta)-\chi_{k,2}(\eta)\partial_\eta\chi_{k,1}(\eta) \tag*{(30)} χ ˉ k ˉ , 1 ( z ) ∂ z χ ˉ k ˉ , 2 ( z ) − χ ˉ k ˉ , 2 ( z ) ∂ z χ ˉ k ˉ , 1 ( z ) = χ k , 1 ( η ) ∂ η χ k , 2 ( η ) − χ k , 2 ( η ) ∂ η χ k , 1 ( η ) (30)
leading to β ˉ k ˉ = β k \bar{\beta}_{\bar{k}}=\beta_k β ˉ k ˉ = β k . To reduce visual clutter in notation, we will from now until a later section eq. (169) drop the underscore on β ˉ k ˉ \bar{\beta}_{\bar{k}} β ˉ k ˉ and the bar on k ˉ \bar{k} k ˉ
β ˉ k ˉ → β k . (31) \bar{\beta}_{\bar{k}}\to\beta_k . \tag*{(31)} β ˉ k ˉ → β k . (31)
The analog of eq. (20) is
χ k ( z ) = α k ( z ) f − ( k , z ) + β k ( z ) f + ( k , z ) (32) \chi_k(z)=\alpha_k(z)f_-(k,z)+\beta_k(z)f_+(k,z) \tag*{(32)} χ k ( z ) = α k ( z ) f − ( k , z ) + β k ( z ) f + ( k , z ) (32)
where α k ( z ) \alpha_k(z) α k ( z ) and β k ( z ) \beta_k(z) β k ( z ) are the complexified WKB amplitudes and f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) are analytically continued WKB basis functions defined as
f ± ( k , z ) ≡ e ± i θ ( k , z ) 2 ω k ( z ) , θ ( k , z ) ≡ ∫ 0 z d z ′ ω k ( z ′ ) . (33) f_\pm(k,z)\equiv\frac{e^{\pm i\theta(k,z)}}{\sqrt{2\omega_k(z)}},\qquad\theta(k,z)\equiv\int_0^z dz'\omega_k(z'). \tag*{(33)} f ± ( k , z ) ≡ 2 ω k ( z ) e ± i θ ( k , z ) , θ ( k , z ) ≡ ∫ 0 z d z ′ ω k ( z ′ ) . (33)
We note that f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) inherit multivaluedness from the fractional powers of ω k 2 \omega_k^2 ω k 2 , such as ω k \omega_k ω k and ω k \sqrt{\omega_k} ω k . Therefore, the decomposition (32) of the complexified WKB amplitudes α k ( z ) \alpha_k(z) α k ( z ) and β k ( z ) \beta_k(z) β k ( z ) over the fictitious WKB basis requires a choice of branch of ω k ( z ) \omega_k(z) ω k ( z ) and ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) . Since on the real line, ω k ( η ) \omega_k(\eta) ω k ( η ) corresponds to the energy of approximate single particle states, we require it to be non-negative and continuous for all η ∈ R \eta\in\mathbb{R} η ∈ R . This implies that the branch cuts defining ω k ( z ) \omega_k(z) ω k ( z ) and ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) , arising from the zeros of ω k 2 \omega_k^2 ω k 2 , have to be taken away from the real line.7 In § 3 and appendix § 8 , we detail this branch choice and its consequences, one of which is that the fictitious WKB basis matches the physical WKB basis on the real line, ensuring that the complexified WKB amplitudes are equal to their real versions on the real line.
With this branch choice, the propagation of the analytically continued WKB amplitudes between two complex points z z z and z 0 z_0 z 0 is
V k ( z ) = U 0 ( k , z , z 0 ) V k ( z 0 ) , V k ( z ) = ( α k ( z ) β k ( z ) ) , (34) V_k(z)=U_0(k,z,z_0)V_k(z_0),\qquad V_k(z)=\begin{pmatrix}\alpha_k(z)\\\beta_k(z)\end{pmatrix}, \tag*{(34)} V k ( z ) = U 0 ( k , z , z 0 ) V k ( z 0 ) , V k ( z ) = ( α k ( z ) β k ( z ) ) , (34)
where the propagator is given by
U 0 ( k , z , z 0 ) = P [ I 2 × 2 exp ( ∫ Γ ( z , z 0 ) d w M 0 ( k , w ) ) ] (35) U_0(k,z,z_0)=\mathbb{P}\left[\mathbb{I}_{2\times2}\exp\left(\int_{\Gamma(z,z_0)}dwM_0(k,w)\right)\right] \tag*{(35)} U 0 ( k , z , z 0 ) = P [ I 2 × 2 exp ( ∫ Γ ( z , z 0 ) d w M 0 ( k , w ) ) ] (35)
M 0 ( k , z ) = ω k ′ ( z ) 2 ω k ( z ) ( 0 e + 2 i ∫ 0 z d z ′ ω k ( z ′ ) e − 2 i ∫ 0 z d z ′ ω k ( z ′ ) 0 ) (36) M_0(k,z)=\frac{\omega'_k(z)}{2\omega_k(z)}
\begin{pmatrix}
0 & e^{+2i\int_0^z dz'\omega_k(z')}\\
e^{-2i\int_0^z dz'\omega_k(z')} & 0
\end{pmatrix} \tag*{(36)} M 0 ( k , z ) = 2 ω k ( z ) ω k ′ ( z ) ( 0 e − 2 i ∫ 0 z d z ′ ω k ( z ′ ) e + 2 i ∫ 0 z d z ′ ω k ( z ′ ) 0 ) (36)
and P \mathbb{P} P denotes path-ordering and Γ ( z , z 0 ) \Gamma(z,z_0) Γ ( z , z 0 ) denotes the integration path connecting z 0 z_0 z 0 and z z z . In terms of these, the Bogoliubov coefficients of eq. (14) are given as the 11 and 21 entries of the propagator from the asymptotic past to the asymptotic future, that is
α k ≡ [ lim z → η + ∞ lim z 0 → η − ∞ U 0 ( k , z , z 0 ) ] 11 (37) \alpha_k \equiv\left[\lim_{z\to\eta_{+\infty}}\lim_{z_0\to\eta_{-\infty}} U_0(k,z,z_0)\right]_{11} \tag*{(37)} α k ≡ [ z → η + ∞ lim z 0 → η − ∞ lim U 0 ( k , z , z 0 ) ] 11 (37)
β k ≡ [ lim z → η + ∞ lim z 0 → η − ∞ U 0 ( k , z , z 0 ) ] 21 . (38) \beta_k \equiv\left[\lim_{z\to\eta_{+\infty}}\lim_{z_0\to\eta_{-\infty}} U_0(k,z,z_0)\right]_{21}. \tag*{(38)} β k ≡ [ z → η + ∞ lim z 0 → η − ∞ lim U 0 ( k , z , z 0 ) ] 21 . (38)
The above expressions provide a precise formulation of tracking the evolution of the complexified WKB modes we mentioned in the beginning of this subsection. The advantage of these expressions, for dispersion relations like eq. (29) , is that the propagation contour Γ ( z , z 0 ) \Gamma(z,z_0) Γ ( z , z 0 ) can be chosen as depicted in Figure 1 such that the propagator along it becomes easier to evaluate. In particular, away from the branch cuts, Γ ( z , z 0 ) \Gamma(z,z_0) Γ ( z , z 0 ) is chosen to lie in the asymptotic annulus (AA) where ∣ z ∣ ≫ k 2 / n |z| \gg k^{2/n} ∣ z ∣ ≫ k 2/ n , whereas near the branch cuts the contour is deformed with keyhole shaped detours.8 In the following, we quantify the propagators along contour segments in the AA and segments near the branch cuts.
Figure 1. The figure depicts the analytic structure associated with the WKB basis functions f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) on the complex plane, for the dispersion relation ω k ( z ) = k 2 + z 4 \omega_k(z)=\sqrt{k^2+z^4} ω k ( z ) = k 2 + z 4 . The zeros of the dispersion relation denoted as ψ i \psi_i ψ i have been shown as blue dots from which wavy lines representing branch cuts arise. The green annulus and the white disk at the center correspond to the regions where the dispersion relation becomes adiabatic and non-adiabatic, respectively; and the blue and red dashed lines to Stokes lines (SLs) and anti-Stokes lines (ASLs). Note, the branch cuts have been taken to lie along the ASLs directed away from the real line. The adiabatic region between any two ASLs corresponds to a Stokes sector — denoted S j \mathcal{S}_j S j , j = 1 , 2 , … , n + 2 j=1,2,\ldots,n+2 j = 1 , 2 , … , n + 2 . The first Stokes sector S 1 \mathcal{S}_1 S 1 is taken to be the sector in the first quadrant with the real line as one of its boundaries; higher j j j s are labelled counterclockwise. Also shown is a representative integration contour Γ ( z , z 0 ) \Gamma(z,z_0) Γ ( z , z 0 ) connecting a point z 0 z_0 z 0 on the negative real axis to a point z z z on the positive real axis. The contour is deformed near the branch cuts with keyhole shaped detours.
Across the Stokes lines (Stokes phenomenon) As noted in the introduction to this section, Stokes phenomenon implies that the WKB amplitude of the subdominant mode undergoes a discrete shift by an O ( 1 ) O(1) O ( 1 ) constant (to O ( z − 1 ) O(z^{-1}) O ( z − 1 ) ) across the Stokes lines. In terms of the propagator in the AA, this shift across one Stokes line (SL) may be represented as [9 , 35 ]
U + ( S ( k ) ) ∼ ( 1 S ( k ) 0 1 ) , U − ( S ( k ) ) ∼ ( 1 0 S ( k ) 1 ) , (39) U_+(\mathcal{S}(k)) \sim\begin{pmatrix} 1 & S(k) \\ 0 & 1 \end{pmatrix}, \qquad U_-(\mathcal{S}(k)) \sim\begin{pmatrix} 1 & 0 \\ S(k) & 1 \end{pmatrix}, \tag*{(39)} U + ( S ( k )) ∼ ( 1 0 S ( k ) 1 ) , U − ( S ( k )) ∼ ( 1 S ( k ) 0 1 ) , (39)
where U ± U_{\pm} U ± refer to the propagators across ± \pm ± SL (defined below eq. (19) ), respectively, and S ( k ) S(k) S ( k ) are the Stokes constants.9 Between any two Stokes lines, there exists an anti-Stokes line (ASL) defined as a contour along which
i θ ( k , z ) ∈ imaginary . (40) i\theta(k,z) \in\text{imaginary}. \tag*{(40)} i θ ( k , z ) ∈ imaginary . (40)
These ASLs separate the complex plane into different Stokes sectors. In the AA, for the given ω k ( z ) = k 2 + z n \omega_k(z)=\sqrt{k^2+z^n} ω k ( z ) = k 2 + z n , the anti-Stokes lines asymptote towards radial lines at the arguments 2 π ( j − 1 ) / ( n + 2 ) 2\pi(j-1)/(n+2) 2 π ( j − 1 ) / ( n + 2 ) and the Stokes lines to those at π ( 2 j − 1 ) / ( n + 2 ) \pi(2j-1)/(n+2) π ( 2 j − 1 ) / ( n + 2 ) , where j = 1 , 2 , ⋯ , n + 2 j=1,2,\cdots,n+2 j = 1 , 2 , ⋯ , n + 2 . We define the asymptotic region between the j j j -th and the ( j + 1 ) (j+1) ( j + 1 ) -th anti-Stokes lines the j j j -th Stokes sector S j \mathcal{S}_j S j , as illustrated in Figure 2 .
Figure 2. Shown is a schematic for branch cuts and Stokes sectors in the complex z z z -plane for the dispersion relation ω k ( z ) = k 2 + z n \omega_k(z) = \sqrt{k^2 + z^n} ω k ( z ) = k 2 + z n . The left and right panels correspond to n = 4 n = 4 n = 4 and n = 6 n = 6 n = 6 , respectively. The wavy blue lines represent the chosen branch cuts emanating from the zeros ψ i \psi_i ψ i and the calligraphic S \mathcal{S} S labels the Stokes sectors.
Across branch cuts To work out the propagator across branch cuts with keyhole shaped contours as shown in Figure 1 , we begin by describing the transformation of the WKB mode functions across a branch cut arising from the j j j -th zero ψ j = k 2 / n e ( 2 j − 1 ) i π / n \psi_j=k^{2/n}e^{(2j-1)i\pi/n} ψ j = k 2/ n e ( 2 j − 1 ) iπ / n of the dispersion relation ω k ( z ) \omega_k(z) ω k ( z ) (29) . The transformation of the WKB amplitudes across the complementary branch cuts is then taken to be inverses of these. Given two points z j ± z_j^\pm z j ± infinitesimally close and padding the branch cut from the j j j -th zero of ω k ( z ) \omega_k(z) ω k ( z ) , the values taken by the dispersion relation at these two points are related as
ω k ( z j + ) = − ω k ( z j − ) . (41) \omega_k(z_j^+)=-\omega_k(z_j^-). \tag*{(41)} ω k ( z j + ) = − ω k ( z j − ) . (41)
This in turn implies that the WKB phases θ ( k , z j ± ) \theta(k,z_j^\pm) θ ( k , z j ± ) are related as
θ ( k , z j + ) = ∫ 0 ψ j d z ω k ( z ) + ∫ ψ j z j + d z ω k ( z ) = 2 Δ j ( k ) − θ ( k , z j − ) (42) \theta(k,z_j^+)=\int_0^{\psi_j} dz\omega_k(z)+\int_{\psi_j}^{z_j^+} dz\omega_k(z)=2\Delta_j(k)-\theta(k,z_j^-) \tag*{(42)} θ ( k , z j + ) = ∫ 0 ψ j d z ω k ( z ) + ∫ ψ j z j + d z ω k ( z ) = 2 Δ j ( k ) − θ ( k , z j − ) (42)
where
Δ j ( k ) ≡ ∫ 0 ψ j d z ω k ( z ) . (43) \Delta_j(k)\equiv\int_0^{\psi_j} dz\omega_k(z). \tag*{(43)} Δ j ( k ) ≡ ∫ 0 ψ j d z ω k ( z ) . (43)
Using this and taking into account the transformation of ω k − 1 / 2 ( z ) \omega_k^{-1/2}(z) ω k − 1/2 ( z ) (see eq. (A.14)) in the denominator of the WKB basis f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) , we have
F k T ( z j + ) = F k T ( z j − ) ( 0 i e 2 i Δ j ( k ) i e − 2 i Δ j ( k ) 0 ) (44) F_k^T(z_j^+)=F_k^T(z_j^-)\begin{pmatrix} 0 & ie^{2i\Delta_j(k)} \\ ie^{-2i\Delta_j(k)} & 0 \end{pmatrix} \tag*{(44)} F k T ( z j + ) = F k T ( z j − ) ( 0 i e − 2 i Δ j ( k ) i e 2 i Δ j ( k ) 0 ) (44)
where F k ( z ) ≡ ( f − ( k , z ) f + ( k , z ) ) T F_k(z)\equiv\left(f_-(k,z)\ f_+(k,z)\right)^T F k ( z ) ≡ ( f − ( k , z ) f + ( k , z ) ) T denotes the WKB basis vector. Writing the exact mode decomposition (32) in vector form χ k ( z ) = F k T ( z ) V k ( z ) \chi_k(z)=F_k^T(z)V_k(z) χ k ( z ) = F k T ( z ) V k ( z ) and noting the smoothness of the exact mode χ k ( z j − ) = χ k ( z j + ) \chi_k(z_j^-)=\chi_k(z_j^+) χ k ( z j − ) = χ k ( z j + ) , we obtain the WKB amplitudes propagation across the j j j -th branch cut:
V k ( z j + ) = B j ( k ) V k ( z j − ) , B j ( k ) ≡ ( 0 − i e 2 i Δ j ( k ) − i e − 2 i Δ j ( k ) 0 ) (45) V_k(z_j^+)=\mathcal{B}_j(k)V_k(z_j^-),\qquad\mathcal{B}_j(k)\equiv\begin{pmatrix} 0 & -ie^{2i\Delta_j(k)} \\ -ie^{-2i\Delta_j(k)} & 0 \end{pmatrix} \tag*{(45)} V k ( z j + ) = B j ( k ) V k ( z j − ) , B j ( k ) ≡ ( 0 − i e − 2 i Δ j ( k ) − i e 2 i Δ j ( k ) 0 ) (45)
where B j ( k ) \mathcal{B}_j(k) B j ( k ) denotes the propagator matrix across the branch cut from the j j j -th zero. With the above, we can now rewrite U 0 ( k , η + ∞ , η − ∞ ) U_0(k,\eta_{+\infty},\eta_{-\infty}) U 0 ( k , η + ∞ , η − ∞ ) appearing in the Bogoliubov β k \beta_k β k computation (38) as a product of propagator matrices across the relevant SLs and branch cuts. For convenience, the branch cuts are further taken to be such that in the adiabatic region these asymptote to the ASLs directed away from the real line. With this choice of the branch cuts, the SLs in the upper half plane are of the “−” type and those of the lower half plane of the “+” type. To see this note, for this branch choice the dispersion relation asymptotes to the form ω k ( z ) ∼ s j z n / 2 \omega_k(z) \sim s_j z^{n/2} ω k ( z ) ∼ s j z n /2 for z ∈ S j z \in\mathcal{S}_j z ∈ S j .10 Therefore, the propagator matrix across each Stokes sector takes the form
U j ( k ) ∼ { ( 1 0 S j ( k ) 1 ) + O ( z − 1 ) 1 ≤ j ≤ n / 2 + 1 ( 1 S j ( k ) 0 1 ) + O ( z − 1 ) n / 2 + 2 ≤ j ≤ n + 2 (46) U_j(k) \sim
\begin{cases}
\begin{pmatrix} 1 & 0 \\ S_j(k) & 1 \end{pmatrix} + O(z^{-1}) & 1 \leq j \leq n/2+1 \\
\begin{pmatrix} 1 & S_j(k) \\ 0 & 1 \end{pmatrix} + O(z^{-1}) & n/2+2 \leq j \leq n+2
\end{cases} \tag*{(46)} U j ( k ) ∼ ⎩ ⎨ ⎧ ( 1 S j ( k ) 0 1 ) + O ( z − 1 ) ( 1 0 S j ( k ) 1 ) + O ( z − 1 ) 1 ≤ j ≤ n /2 + 1 n /2 + 2 ≤ j ≤ n + 2 (46)
Combining eqs. (46) and (45) for the contour shown in Figure 1 , we obtain the asymptotic expression for the propagator from the asymptotic past to the asymptotic future
lim z → η + i ∞ lim z 0 → η − i ∞ U 0 ( k , z , z 0 ) = U 2 N + 2 B 2 N U 2 N + 1 ⋯ B N + 1 U N + 2 , (47) \lim_{z\to\eta+i\infty}\lim_{z_0\to\eta-i\infty} U_0(k,z,z_0) = U_{2N+2}\mathcal{B}_{2N}U_{2N+1}\cdots\mathcal{B}_{N+1}U_{N+2}, \tag*{(47)} z → η + i ∞ lim z 0 → η − i ∞ lim U 0 ( k , z , z 0 ) = U 2 N + 2 B 2 N U 2 N + 1 ⋯ B N + 1 U N + 2 , (47)
where we have suppressed the k k k dependence in the right hand side and introduced N ≡ n / 2 N \equiv n/2 N ≡ n /2 . From eq. (38) , we then get
β k = [ U 2 N + 2 B 2 N U 2 N + 1 ⋯ B N + 1 U N + 2 ] 21 (48) \beta_k = \left[U_{2N+2}\mathcal{B}_{2N}U_{2N+1}\cdots\mathcal{B}_{N+1}U_{N+2}\right]_{21} \tag*{(48)} β k = [ U 2 N + 2 B 2 N U 2 N + 1 ⋯ B N + 1 U N + 2 ] 21 (48)
with the subscript 21 denoting the 21 entry of the matrix product. We see that β k \beta_k β k is now expressed in terms of the Stokes constants S j ( k ) S_j(k) S j ( k ) and the non-analytic phase factors Δ j ( k ) \Delta_j(k) Δ j ( k ) . To obtain the soft corrections to the Bogoliubov β k \beta_k β k , what we need are the expressions for Stokes constants S j ( k ) S_j(k) S j ( k ) for j = 1 , … , 2 N + 2 j=1,\ldots,2N+2 j = 1 , … , 2 N + 2 , which are typically very complicated for nonzero momentum. To overcome this complication, in section 3, we will first establish representations of a discrete rotational symmetry of the mode equation (29) on the propagators, which allows us to express all Stokes constants in terms of the first Stokes constant S 1 S_1 S 1 . We also show that a time-reversal symmetry for the real variable mode equation (8) implies that the Bogoliubov β k \beta_k β k must be purely imaginary.
Discrete symmetries
In the topological limit case (k = 0 k=0 k = 0 ), as noted in [9 ] , the mode equation (29) is invariant under the rotation z → γ z z \to\gamma z z → γ z where γ ≡ exp ( 2 π i / ( n + 2 ) ) \gamma\equiv\exp(2\pi i/(n+2)) γ ≡ exp ( 2 π i / ( n + 2 )) . This invariance induces a symmetry representation on the set of propagators across SLs (see eq. (46) ), equating all Stokes constants in the system. The single Stokes constant can then be determined using the single-valuedness of the exact mode χ k ( z ) \chi_k(z) χ k ( z ) upon propagation by 2 π 2\pi 2 π on a circle.
For k ≠ 0 k \ne0 k = 0 , we hope to extend the above argument to relate the different Stokes constants of the mode equation. Note, under z → γ z z \to\gamma z z → γ z the mode equation (29) transforms as
[ ∂ z 2 + ( γ − n / 2 k ) 2 + z n ] χ k ( γ z ) = 0 (49) \left[\partial_z^2 + \left(\gamma^{-n/2}k\right)^2 + z^n\right]\chi_k(\gamma z) = 0 \tag*{(49)} [ ∂ z 2 + ( γ − n /2 k ) 2 + z n ] χ k ( γ z ) = 0 (49)
where χ k ( z ) \chi_k(z) χ k ( z ) is a solution to the unrotated mode equation (29) . This implies a relation between the Stokes constants of eq. (2.27, which are functions of k k k , and the Stokes constants of eq. (49) , which are functions of γ − n / 2 k \gamma^{-n/2}k γ − n /2 k . We derive this relation in § 3.1 which relates the different Stokes constants of eq. (29) to analytic continuations in k k k of a single function S 1 ( k ) S_1(k) S 1 ( k ) . This function is further constrained by the single-valuedness of the mode function, but as we will see in appendix § 11 , this is insufficient to constrain the Stokes constants due to a redundancy in the type of functions which can solve this constraint. Nevertheless, the relation among all Stokes constants reduces the problem to computing just S 1 ( k ) S_1(k) S 1 ( k ) , which we turn to in § 4 .
In addition to the above, restricting to real conformal time, the mode equation (29) is symmetric under the time-reversal transformation { η → − η } \{\eta\to-\eta\} { η → − η } . This induces another symmetry representation on the solution space S \mathcal{S} S of eq. (29) , yielding a constraint on the Bogoliubov β k \beta_k β k . In § 3.2 , we construct this representation and use the Wronskian projection (16) to prove that the Bogoliubov β k \beta_k β k must be imaginary, providing a simple consistency check for our analysis in § 4 .
Rotational symmetry
Here, we show that the propagators across the different SLs corresponding to eq. (29) , as a consequence of the latter’s symmetry under { z → γ z , k → γ − 1 k } \{z \to\gamma z, k \to\gamma^{-1}k\} { z → γ z , k → γ − 1 k } , can be related to analytic continuations of the single function S 1 ( k ) S_1(k) S 1 ( k ) .
We begin by working out a more transparent representation of ω k ( z ) \omega_k(z) ω k ( z ) given the branch cut choice depicted in Figure 1 (see appendix § 8 for a more detailed version). As depicted in Figure 3 , we define branch sectors B l \mathcal{B}_l B l as the region bounded between the branch cuts arising from ψ ( l − 1 ) m o d n \psi_{(l-1)\bmod n} ψ ( l − 1 ) mod n and ψ l m o d n \psi_{l\bmod n} ψ l mod n for ∣ z ∣ > k 2 / n |z| > k^{2/n} ∣ z ∣ > k 2/ n .11 Across these regions, the analytic expression for ω k ( z ) \omega_k(z) ω k ( z ) changes discretely by overall factors of − 1 -1 − 1 corresponding to the two roots ± ω k 2 ( z ) \pm\sqrt{\omega_k^2(z)} ± ω k 2 ( z ) .12 More precisely, in terms of the following Taylor series ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) defined to be analytic in the annulus ∣ z ∣ > k 2 / n |z| > k^{2/n} ∣ z ∣ > k 2/ n
Figure 3. Branch cut structure of ω k ( z ) \omega_k(z) ω k ( z ) is shown schematically for ω k ( z ) = k 2 + z 4 \omega_k(z) = \sqrt{k^2 + z^4} ω k ( z ) = k 2 + z 4 . The branch cuts are chosen to be coincident with anti-Stokes boundaries that are radially emanating in the AA shown in green. In contrast, the basis function ω k ( z ) \omega_k(z) ω k ( z ) of eq. (50) has no branch cuts and is analytic for ∣ z ∣ > ∣ ψ j ∣ |z| > |\psi_j| ∣ z ∣ > ∣ ψ j ∣ where ψ j \psi_j ψ j are the branch points of ω k ( z ) \omega_k(z) ω k ( z ) . We define the branch sectors B j \mathcal{B}_j B j as the region bounded by two branch cuts exterior to the disk ∣ z ∣ ≤ ∣ ψ j ∣ |z| \leq|\psi_j| ∣ z ∣ ≤ ∣ ψ j ∣ . The inner annular sector (orange) is not an asymptotic expansion region, but ω k ( z ) \omega_k(z) ω k ( z ) is analytic there as well. Also, the anti-Stokes lines and the attendant branch cuts are not as straight as schematically shown here.
ω ‾ k ( z ) ≡ z n / 2 ∑ m = 0 ∞ ( 1 / 2 m ) ( k 2 z n ) m (50) \underline{\omega}_k(z) \equiv z^{n/2} \sum_{m=0}^{\infty} \binom{1/2}{m} \left(\frac{k^2}{z^n}\right)^m \tag*{(50)} ω k ( z ) ≡ z n /2 m = 0 ∑ ∞ ( m 1/2 ) ( z n k 2 ) m (50)
ω k ( z ) \omega_k(z) ω k ( z ) in B l \mathcal{B}_l B l for the above branch choice is given by
ω k ( z ) = ( − 1 ) l − 1 ω ‾ k ( z ) , z ∈ B l (51) \omega_k(z) = (-1)^{l-1}\underline{\omega}_k(z), \qquad z \in\mathcal{B}_l \tag*{(51)} ω k ( z ) = ( − 1 ) l − 1 ω k ( z ) , z ∈ B l (51)
The choice of eq. (51) is such that for real time η \eta η and nonzero momentum, for any even integer n n n , ω k ( η ) > 0 \omega_k(\eta) > 0 ω k ( η ) > 0 . Correspondingly, in the inner disk D D D defined as ∣ z ∣ ≤ k 2 / n |z| \le k^{2/n} ∣ z ∣ ≤ k 2/ n , ω k ( z ) \omega_k(z) ω k ( z ) is given by the root whose restriction on the real line is positive definite, denoted + k 2 + z n +\sqrt{k^2+z^n} + k 2 + z n . Combining these we have
ω k ( z ) = { + k 2 + z n z ∈ D ( − 1 ) l − 1 ω k ( z ) z ∈ B l (52) \omega_k(z) =
\begin{cases}
+\sqrt{k^2+z^n} & z \in D \\
(-1)^{l-1}\omega_k(z) & z \in\mathcal{B}_l
\end{cases} \tag*{(52)} ω k ( z ) = { + k 2 + z n ( − 1 ) l − 1 ω k ( z ) z ∈ D z ∈ B l (52)
Note the Stokes sectors S j \mathcal{S}_j S j are subsets of corresponding branch sectors. With our branch choice, eq. (52) can be expressed in terms of Stokes sector labels
ω k ( z ) = { + k 2 + z n z ∈ D s j ω k ( z ) z ∈ S j E (53) \omega_k(z) =
\begin{cases}
+\sqrt{k^2+z^n} & z \in D \\
s_j\omega_k(z) & z \in\mathcal{S}_j^E
\end{cases} \tag*{(53)} ω k ( z ) = { + k 2 + z n s j ω k ( z ) z ∈ D z ∈ S j E (53)
where
s j ≡ { ( − 1 ) j − 1 1 ≤ j ≤ n / 2 + 1 ( − 1 ) j n / 2 + 2 ≤ j ≤ n + 2 (54) s_j \equiv
\begin{cases}
(-1)^{j-1} & 1 \leq j \leq n/2+1 \\
(-1)^j & n/2+2 \leq j \leq n+2
\end{cases} \tag*{(54)} s j ≡ { ( − 1 ) j − 1 ( − 1 ) j 1 ≤ j ≤ n /2 + 1 n /2 + 2 ≤ j ≤ n + 2 (54)
and where the superscript “E” in S j E \mathcal{S}_j^E S j E corresponds to the extension of the Stokes sector to cover the region between the Stokes sector and the inner disk.13
S j E = { ∣ z ∣ > ∣ ψ j ∣ ∣ z restricted to the region bounded by anti-Stokes boundaries belonging to S j } \mathcal{S}_j^E = \{ |z| > |\psi_j| \mid z\ \text{restricted to the region bounded by anti-Stokes boundaries belonging to } \mathcal{S}_j \} S j E = { ∣ z ∣ > ∣ ψ j ∣ ∣ z restricted to the region bounded by anti-Stokes boundaries belonging to S j }
where we remind the reader that the branch cut has been chosen to be coincident with the outward
With the above, consider the j j j -th Stokes constant given by
S j ( k ) = lim ∣ z j 0 ∣ , ∣ z j 1 ∣ → ∞ [ U j ( k ) ] 21 ( 12 ) (55) S_j(k)=\lim_{\lvert z_{j0}\rvert,\lvert z_{j1}\rvert\to\infty}[U_j(k)]_{21(12)} \tag*{(55)} S j ( k ) = ∣ z j 0 ∣ , ∣ z j 1 ∣ → ∞ lim [ U j ( k ) ] 21 ( 12 ) (55)
where z j 1 z_{j1} z j 1 and z j 0 z_{j0} z j 0 are two points on the ASLs bounding the j j j -th Stokes sector S j \mathcal{S}_j S j — avoiding the branch cuts,14 and where
U j ( k ) ≡ U 0 ( k , z j 1 , z j 0 ) = P [ I 2 × 2 exp ( ∫ Γ ( z j 1 , z j 0 ) d w M 0 ( k , w ) ) ] (56) U_j(k)\equiv U_0(k,z_{j1},z_{j0})=\mathbb{P}\left[\mathbb{I}_{2\times2}\exp\left(\int_{\Gamma(z_{j1},z_{j0})}dwM_0(k,w)\right)\right] \tag*{(56)} U j ( k ) ≡ U 0 ( k , z j 1 , z j 0 ) = P [ I 2 × 2 exp ( ∫ Γ ( z j 1 , z j 0 ) d w M 0 ( k , w ) ) ] (56)
M 0 ( k , z ) = ω k ′ ( z ) 2 ω k ( z ) ( 0 e + 2 i θ ( k , z ) e − 2 i θ ( k , z ) 0 ) , θ ( k , z ) ≡ ∫ 0 z d z ′ ω k ( z ′ ) . (57) M_0(k,z)=\frac{\omega_k'(z)}{2\omega_k(z)}
\begin{pmatrix}
0 & e^{+2i\theta(k,z)}\\
e^{-2i\theta(k,z)} & 0
\end{pmatrix},
\qquad
\theta(k,z)\equiv\int_0^z dz'\omega_k(z'). \tag*{(57)} M 0 ( k , z ) = 2 ω k ( z ) ω k ′ ( z ) ( 0 e − 2 i θ ( k , z ) e + 2 i θ ( k , z ) 0 ) , θ ( k , z ) ≡ ∫ 0 z d z ′ ω k ( z ′ ) . (57)
To derive the symmetry representation on the Stokes constants as functions of k k k , we need to consider the rotation z → γ z z\to\gamma z z → γ z , as mentioned above eq. (49) , in the WKB phase θ ( k , z ) \theta(k,z) θ ( k , z ) in the propagators. However, this rotation is obstructed when there is a branch cut between z z z and γ z \gamma z γ z , as is the case when rotating between two Stokes sectors S j \mathcal{S}_j S j and S j + 1 \mathcal{S}_{j+1} S j + 1 separated by the j j j -th branch cut. Nevertheless, for this case, we may introduce an auxiliary dispersion relation ω k I I [ j ] ( z ) \omega_k^{\mathrm{II}[j]}(z) ω k II [ j ] ( z ) that lands on the second Riemann sheet in S j + 1 E \mathcal{S}_{j+1}^{\mathrm{E}} S j + 1 E which equals ω k ( z ) \omega_k(z) ω k ( z ) otherwise. This construction implies that ω k I I [ j ] ( z ) \omega_k^{\mathrm{II}[j]}(z) ω k II [ j ] ( z ) is continuous across the j j j -th branch cut of ω k ( z ) \omega_k(z) ω k ( z ) , which makes rotating the WKB phase of ω k I I [ j ] ( z ) \omega_k^{\mathrm{II}[j]}(z) ω k II [ j ] ( z ) more transparent. Formally, we define
ω k I I [ j ] ( z ) ≡ { − ω k ( z ) z ∈ S j + 1 E + ω k ( z ) otherwise , b = 1 , (58) \omega_k^{\mathrm{II}[j]}(z)\equiv
\begin{cases}
-\omega_k(z) & z\in\mathcal{S}_{j+1}^{\mathrm{E}}\\
+\omega_k(z) & \text{otherwise}
\end{cases},
\qquad b=1, \tag*{(58)} ω k II [ j ] ( z ) ≡ { − ω k ( z ) + ω k ( z ) z ∈ S j + 1 E otherwise , b = 1 , (58)
where b ∈ { 0 , 1 } b\in\{0,1\} b ∈ { 0 , 1 } denotes the number of branch cuts between S j \mathcal{S}_j S j and S j + 1 \mathcal{S}_{j+1} S j + 1 . Compared to the original WKB phase θ ( k , z ) \theta(k,z) θ ( k , z ) , the phase integral of the auxiliary dispersion relation in S j \mathcal{S}_j S j and S j + 1 \mathcal{S}_{j+1} S j + 1 satisfies
θ I I [ j ] ( k , z ) ≡ ∫ 0 z d z ′ ω k I I [ j ] ( z ′ ) = { θ ( k , z ) z ∈ S j 2 Δ j ( k ) − θ ( k , z ) z ∈ S j + 1 , b = 1 , (59) \theta^{\mathrm{II}[j]}(k,z)\equiv\int_0^z dz'\omega_k^{\mathrm{II}[j]}(z')=
\begin{cases}
\theta(k,z) & z\in\mathcal{S}_j\\
2\Delta_j(k)-\theta(k,z) & z\in\mathcal{S}_{j+1}
\end{cases},
\qquad b=1, \tag*{(59)} θ II [ j ] ( k , z ) ≡ ∫ 0 z d z ′ ω k II [ j ] ( z ′ ) = { θ ( k , z ) 2 Δ j ( k ) − θ ( k , z ) z ∈ S j z ∈ S j + 1 , b = 1 , (59)
where the phase factor Δ j ( k ) \Delta_j(k) Δ j ( k ) is defined in eq. (43) . Consequently, in the following where we define analytic continuations of the functions S j ( k ) S_j(k) S j ( k ) to complex momenta p = γ − n m / 2 k p=\gamma^{-nm/2}k p = γ − nm /2 k where m ∈ Z ≥ 0 m\in\mathbb{Z}_{\geq0} m ∈ Z ≥ 0 , the corresponding dispersion relation ω p ( z ) \omega_p(z) ω p ( z ) is taken to have branch cuts such that for ∣ z ∣ ≫ ∣ p ∣ 2 / n \lvert z\rvert\gg\lvert p\rvert^{2/n} ∣ z ∣ ≫ ∣ p ∣ 2/ n , ω p ( z ) / ω k ( z ) ∼ 1 \omega_p(z)/\omega_k(z)\sim1 ω p ( z ) / ω k ( z ) ∼ 1 . Therefore, for this momentum complexification we have
ω p ( z ) = s j ω p ( z ) , z ∈ S j , (60) \omega_p(z)=s_j\omega_p(z),\qquad z\in\mathcal{S}_j, \tag*{(60)} ω p ( z ) = s j ω p ( z ) , z ∈ S j , (60)
emanating anti-Stokes boundaries. Each real axis sliver between each set of two anti-Stokes boundaries surrounding the real axis can be evenly included in the surrounding S j E \mathcal{S}_j^{\mathrm{E}} S j E to not leave out any ∣ z ∣ > ∣ ψ j ∣ \lvert z\rvert>\lvert\psi_j\rvert ∣ z ∣ > ∣ ψ j ∣ .
z j 0 ≡ R exp [ i ( 2 π ( j − 1 ) n + 2 + ϵ ) ] , z j 1 ≡ R exp [ i ( 2 π j n + 2 − ϵ ) ] z_{j0}\equiv R\exp\left[i\left(\frac{2\pi(j-1)}{n+2}+\epsilon\right)\right],
\qquad
z_{j1}\equiv R\exp\left[i\left(\frac{2\pi j}{n+2}-\epsilon\right)\right] z j 0 ≡ R exp [ i ( n + 2 2 π ( j − 1 ) + ϵ ) ] , z j 1 ≡ R exp [ i ( n + 2 2 π j − ϵ ) ]
where R R R is a large real number and ϵ → 0 + \epsilon\to0^+ ϵ → 0 + a positive infinitesimal regulator.
where ω p ( z ) \omega_p(z) ω p ( z ) denotes the Taylor series (50) with k k k replaced by p p p . Such a branch choice, shown in Figure 4 , can be obtained by replacing k k k with p p p in eq. (50) and analytically continuing to z ∈ D z \in D z ∈ D . This complexifies the momentum in the WKB phase as θ ( p , z ) ≡ ∫ 0 z d z ′ ω p ( z ′ ) \theta(p,z) \equiv\int_0^z \mathrm{d}z'\omega_p(z') θ ( p , z ) ≡ ∫ 0 z d z ′ ω p ( z ′ ) .15 Note this preserves the Stokes sectors S j \mathcal{S}_j S j for the case of continued momenta.
Figure 4. Shown is our convention for branch cut deformation under the complexification k → p k \to p k → p for n = 4 n = 4 n = 4 . As arg p \arg p arg p varies, the branch points ψ j ( p ) \psi_j(p) ψ j ( p ) rotate and drag the branch cuts emanating from them. We choose to deform the cuts only in the annulus while keeping their outer segments fixed. The corresponding branch sectors, and hence the factors s j s_j s j , may therefore be kept fixed. This guarantees that ω p ( z ) \omega_p(z) ω p ( z ) and ω k ( z ) \omega_k(z) ω k ( z ) have the same leading asymptotic term s j z n / 2 s_j z^{n/2} s j z n /2 for z ∈ S j z \in\mathcal{S}_j z ∈ S j .
We now turn to deriving the symmetry representation of the propagators. Consider two sets of points belonging to adjacent Stokes sectors, given by { γ z j 1 , γ z j 0 } ⊂ S j + 1 \{\gamma z_{j1},\gamma z_{j0}\} \subset\mathcal{S}_{j+1} { γ z j 1 , γ z j 0 } ⊂ S j + 1 and { z j 1 , z j 0 } ⊂ S j \{z_{j1},z_{j0}\} \subset\mathcal{S}_j { z j 1 , z j 0 } ⊂ S j , where these points lie close to the ASLs in each sector while avoiding branch cuts. Given this, the propagator in S j + 1 \mathcal{S}_{j+1} S j + 1 is given by
U j + 1 ( k ) = P [ I 2 × 2 exp ( ∫ Γ ( γ z j 1 , γ z j 0 ) d z M 0 ( k , z ) ) ] = P [ I 2 × 2 exp ( ∫ Γ ( γ z j 1 , γ z j 0 ) d z ω k ′ ( z ) 2 ω k ( z ) ( 0 e + 2 i θ ( k , z ) e − 2 i θ ( k , z ) 0 ) ) ] . (61) \begin{aligned}
U_{j+1}(k)=\mathbb{P}\left[\mathbb{I}_{2\times2}\exp\left(\int_{\Gamma(\gamma z_{j1},\gamma z_{j0})} dz\,M_0(k,z)\right)\right] \\
&=\mathbb{P}\left[\mathbb{I}_{2\times2}\exp\left(\int_{\Gamma(\gamma z_{j1},\gamma z_{j0})} dz\,\frac{\omega_k'(z)}{2\omega_k(z)}
\begin{pmatrix}
0 & e^{+2i\theta(k,z)}\\
e^{-2i\theta(k,z)} & 0
\end{pmatrix}\right)\right]. \tag*{(61)}
\end{aligned} U j + 1 ( k ) = P [ I 2 × 2 exp ( ∫ Γ ( γ z j 1 , γ z j 0 ) d z M 0 ( k , z ) ) ] = P [ I 2 × 2 exp ( ∫ Γ ( γ z j 1 , γ z j 0 ) d z 2 ω k ( z ) ω k ′ ( z ) ( 0 e − 2 i θ ( k , z ) e + 2 i θ ( k , z ) 0 ) ) ] . (61)
where Γ ( γ z j 1 , γ z j 0 ) \Gamma(\gamma z_{j1},\gamma z_{j0}) Γ ( γ z j 1 , γ z j 0 ) denotes the propagation contour.
Under a variable change z = γ y z=\gamma y z = γ y , the exponent of the above path ordered exponential integral goes to
∫ Γ ( γ z j 1 , γ z j 0 ) d z ω k ′ ( z ) 2 ω k ( z ) ( 0 e + 2 i θ ( k , z ) e − 2 i θ ( k , z ) 0 ) = ∫ Γ ( z j 1 , z j 0 ) d y n y n − 1 4 ( γ − n k 2 + y n ) ( 0 e + 2 i θ ( k , γ y ) e − 2 i θ ( k , γ y ) 0 ) . (62) \begin{aligned}
\int_{\Gamma(\gamma z_{j1},\gamma z_{j0})} dz\,\frac{\omega_k'(z)}{2\omega_k(z)}
\begin{pmatrix}
0 & e^{+2i\theta(k,z)}\\
e^{-2i\theta(k,z)} & 0
\end{pmatrix} \\
&=\int_{\Gamma(z_{j1},z_{j0})}dy\,\frac{ny^{n-1}}{4\left(\gamma^{-n}k^2+y^n\right)}
\begin{pmatrix}
0 & e^{+2i\theta(k,\gamma y)}\\
e^{-2i\theta(k,\gamma y)} & 0
\end{pmatrix}. \tag*{(62)}
\end{aligned} ∫ Γ ( γ z j 1 , γ z j 0 ) d z 2 ω k ( z ) ω k ′ ( z ) ( 0 e − 2 i θ ( k , z ) e + 2 i θ ( k , z ) 0 ) = ∫ Γ ( z j 1 , z j 0 ) d y 4 ( γ − n k 2 + y n ) n y n − 1 ( 0 e − 2 i θ ( k , γ y ) e + 2 i θ ( k , γ y ) 0 ) . (62)
where Γ ( z j 1 , z j 0 ) \Gamma(z_{j1},z_{j0}) Γ ( z j 1 , z j 0 ) now lies in S j \mathcal{S}_j S j and note the pre-factor ω k ′ / ω k \omega_k'/\omega_k ω k ′ / ω k is single valued. Now, since γ y \gamma y γ y still lies in S j + 1 \mathcal{S}_{j+1} S j + 1 , we use another variable change z ′ = γ u z'=\gamma u z ′ = γ u in θ ( k , γ y ) = ∫ 0 γ y d z ′ ω k ( z ′ ) \theta(k,\gamma y)=\int_0^{\gamma y} dz'\,\omega_k(z') θ ( k , γ y ) = ∫ 0 γ y d z ′ ω k ( z ′ ) to relate the phase integral θ ( k , γ y ) \theta(k,\gamma y) θ ( k , γ y ) to the corresponding integrals for ω γ − n / 2 k \omega_{\gamma^{-n/2}k} ω γ − n /2 k in S j \mathcal{S}_j S j . If the Stokes sectors are not separated by branch cuts, then this simple variable change leads to the relation
θ ( k , γ y ) = ∫ 0 γ y d z ′ ω k ( z ′ ) = ∫ 0 y γ d u k 2 + γ n u n = ∫ 0 y γ 1 + n / 2 d u γ − n k 2 + u n = − θ ( γ − n / 2 k , y ) , (63) \begin{aligned}
\theta(k,\gamma y)=\int_0^{\gamma y}dz'\,\omega_k(z') \\
&=\int_0^y\gamma\,du\sqrt{k^2+\gamma^n u^n} \\
&=\int_0^y\gamma^{1+n/2}du\sqrt{\gamma^{-n}k^2+u^n} \\
&=-\theta(\gamma^{-n/2}k,y), \tag*{(63)}
\end{aligned} θ ( k , γ y ) = ∫ 0 γ y d z ′ ω k ( z ′ ) = ∫ 0 y γ d u k 2 + γ n u n = ∫ 0 y γ 1 + n /2 d u γ − n k 2 + u n = − θ ( γ − n /2 k , y ) , (63)
where we have used γ 1 + n / 2 = − 1 \gamma^{1+n/2}=-1 γ 1 + n /2 = − 1 . If there is a branch cut of ω k ( z ′ ) \omega_k(z') ω k ( z ′ ) between S j \mathcal{S}_j S j and S j + 1 \mathcal{S}_{j+1} S j + 1 , then we may first use the auxiliary dispersion ω k I I [ j ] ( z ′ ) \omega_k^{\mathrm{II}[j]}(z') ω k II [ j ] ( z ′ ) which is continuous across this branch cut to get the analogue of eq. (63) :
θ I I [ j ] ( k , γ y ) = − θ I I [ j ] ( γ − n / 2 k , y ) . (64) \theta^{\mathrm{II}[j]}(k,\gamma y)=-\theta^{\mathrm{II}[j]}(\gamma^{-n/2}k,y). \tag*{(64)} θ II [ j ] ( k , γ y ) = − θ II [ j ] ( γ − n /2 k , y ) . (64)
Then, we map the above auxiliary WKB phase to θ ( k , z ) \theta(k,z) θ ( k , z ) using eq. (59) , yielding
θ ( k , γ y ) = 2 Δ j ( k ) + θ ( γ − n / 2 k , y ) . (65) \theta(k,\gamma y)=2\Delta_j(k)+\theta(\gamma^{-n/2}k,y). \tag*{(65)} θ ( k , γ y ) = 2 Δ j ( k ) + θ ( γ − n /2 k , y ) . (65)
Combining eqs. (63) and (65) leads to
θ ( k , γ y ) = { 2 Δ j ( k ) + θ ( γ − n / 2 k , y ) b = 1 − θ ( γ − n / 2 k , y ) b = 0 , y ∈ S j . (66) \theta(k,\gamma y)=
\begin{cases}
2\Delta_j(k)+\theta(\gamma^{-n/2}k,y) & b=1\\
-\theta(\gamma^{-n/2}k,y) & b=0
\end{cases},
\qquad y\in\mathcal{S}_j. \tag*{(66)} θ ( k , γ y ) = { 2 Δ j ( k ) + θ ( γ − n /2 k , y ) − θ ( γ − n /2 k , y ) b = 1 b = 0 , y ∈ S j . (66)
This with eq. (57) leads to
∫ Γ ( z j 1 , z j 0 ) γ d y M 0 ( k , γ y ) = { R j ( k ) [ ∫ Γ ( z j 1 , z j 0 ) d y M 0 ( γ − n / 2 k , y ) ] R j − 1 ( k ) b = 1 , B [ ∫ Γ ( z j 1 , z j 0 ) d y M 0 ( γ − n / 2 k , y ) ] B b = 0. (67) \int_{\Gamma(z_{j1},z_{j0})} \gamma\, \mathrm{d}y M_0(k,\gamma y) =
\begin{cases}
\mathcal{R}_j(k)\left[\int_{\Gamma(z_{j1},z_{j0})} \mathrm{d}y M_0(\gamma^{-n/2}k,y)\right]\mathcal{R}_j^{-1}(k) & b=1,\\
B\left[\int_{\Gamma(z_{j1},z_{j0})} \mathrm{d}y M_0(\gamma^{-n/2}k,y)\right]B & b=0.
\end{cases} \tag*{(67)} ∫ Γ ( z j 1 , z j 0 ) γ d y M 0 ( k , γ y ) = ⎩ ⎨ ⎧ R j ( k ) [ ∫ Γ ( z j 1 , z j 0 ) d y M 0 ( γ − n /2 k , y ) ] R j − 1 ( k ) B [ ∫ Γ ( z j 1 , z j 0 ) d y M 0 ( γ − n /2 k , y ) ] B b = 1 , b = 0. (67)
where we have used eq. (66) and introduced
B ≡ ( 0 1 1 0 ) , R j ( k ) ≡ ( e 2 i Δ j ( k ) 0 0 e − 2 i Δ j ( k ) ) . (68) B \equiv\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, \qquad\mathcal{R}_j(k) \equiv\begin{pmatrix} e^{2i\Delta_j(k)} & 0 \\ 0 & e^{-2i\Delta_j(k)} \end{pmatrix}. \tag*{(68)} B ≡ ( 0 1 1 0 ) , R j ( k ) ≡ ( e 2 i Δ j ( k ) 0 0 e − 2 i Δ j ( k ) ) . (68)
This yields a rotational symmetry representation on the propagator in the zero gauge
U j + 1 ( k ) = { R j ( k ) U j ( γ − n / 2 k ) R j − 1 ( k ) 1 ≤ j ≤ n / 2 B U j ( γ − n / 2 k ) B − 1 j = n / 2 + 1 R j − 1 ( k ) U j ( γ − n / 2 k ) R j − 1 − 1 ( k ) n / 2 + 2 ≤ j ≤ n + 1. (69) U_{j+1}(k) =
\begin{cases}
\mathcal{R}_j(k)U_j(\gamma^{-n/2}k)\mathcal{R}_j^{-1}(k) & 1 \le j \le n/2\\
BU_j(\gamma^{-n/2}k)B^{-1} & j=n/2+1\\
\mathcal{R}_{j-1}(k)U_j(\gamma^{-n/2}k)\mathcal{R}_{j-1}^{-1}(k) & n/2+2 \le j \le n+1.
\end{cases} \tag*{(69)} U j + 1 ( k ) = ⎩ ⎨ ⎧ R j ( k ) U j ( γ − n /2 k ) R j − 1 ( k ) B U j ( γ − n /2 k ) B − 1 R j − 1 ( k ) U j ( γ − n /2 k ) R j − 1 − 1 ( k ) 1 ≤ j ≤ n /2 j = n /2 + 1 n /2 + 2 ≤ j ≤ n + 1. (69)
Composing the above j j j times, it then follows from eqs. (35) , (69) and (74) that (see appendix C)
S j + 1 ( k ) = e − 4 i Δ ~ j ( k ) S 1 ( γ − N j k ) , 1 ≤ j ≤ N , (70) S_{j+1}(k)=e^{-4i\widetilde{\Delta}_j(k)}S_1(\gamma^{-Nj}k), \qquad1\le j\le N, \tag*{(70)} S j + 1 ( k ) = e − 4 i Δ j ( k ) S 1 ( γ − N j k ) , 1 ≤ j ≤ N , (70)
S N + 2 ( k ) = e 4 i Δ ~ N ( k ) S 1 ( γ − N ( N + 1 ) k ) , (71) S_{N+2}(k)=e^{4i\widetilde{\Delta}_N(k)}S_1(\gamma^{-N(N+1)}k), \tag*{(71)} S N + 2 ( k ) = e 4 i Δ N ( k ) S 1 ( γ − N ( N + 1 ) k ) , (71)
S N + 2 + j ( k ) = e 4 i Φ ~ j ( k ) S 1 ( γ − N ( N + 1 + j ) k ) , 1 ≤ j ≤ N , (72) S_{N+2+j}(k)=e^{4i\widetilde{\Phi}_j(k)}S_1(\gamma^{-N(N+1+j)}k), \qquad1\le j\le N, \tag*{(72)} S N + 2 + j ( k ) = e 4 i Φ j ( k ) S 1 ( γ − N ( N + 1 + j ) k ) , 1 ≤ j ≤ N , (72)
where Δ ~ j \widetilde{\Delta}_j Δ j and Φ ~ j \widetilde{\Phi}_j Φ j correspond to the following alternating sums
Δ ~ j ( k ) ≡ ∑ r = 1 j ( − 1 ) j − r Δ r ( k ) , Φ ~ j ( k ) = ( − 1 ) j Δ ~ N ( k ) − Δ ~ j ( k ) . (73) \widetilde{\Delta}_j(k)\equiv\sum_{r=1}^{j}(-1)^{j-r}\Delta_r(k), \qquad\widetilde{\Phi}_j(k)=(-1)^j\widetilde{\Delta}_N(k)-\widetilde{\Delta}_j(k). \tag*{(73)} Δ j ( k ) ≡ r = 1 ∑ j ( − 1 ) j − r Δ r ( k ) , Φ j ( k ) = ( − 1 ) j Δ N ( k ) − Δ j ( k ) . (73)
To arrive at the above we have used
Δ j ( γ − N k ) = − Δ j ( k ) , Δ N + j ( k ) = − Δ j ( k ) (74) \Delta_j(\gamma^{-N}k)=-\Delta_j(k), \qquad\Delta_{N+j}(k)=-\Delta_j(k) \tag*{(74)} Δ j ( γ − N k ) = − Δ j ( k ) , Δ N + j ( k ) = − Δ j ( k ) (74)
which follows from Δ j ( k ) ∝ k n + 2 n \Delta_j(k)\propto k^{\frac{n+2}{n}} Δ j ( k ) ∝ k n n + 2 as may be derived from the definition (43) of Δ j ( k ) \Delta_j(k) Δ j ( k ) with the variable change z = u k 2 / n z=uk^{2/n} z = u k 2/ n . The three relations (70) -(72) express all the Stokes constants in terms of the first Stokes constant and certain phase factors. As a result, to compute the Bogoliubov β k \beta_k β k using propagator with eq. (48) , we only need to compute the first Stokes constant S 1 ( k ) S_1(k) S 1 ( k ) , which will be the task for section 4.
Time-reversal symmetry
The even parity of n n n guarantees that the dispersion relation ω k ( η ) = k 2 + η n \omega_k(\eta)=\sqrt{k^2+\eta^n} ω k ( η ) = k 2 + η n is invariant under time-reversal η → − η \eta\to-\eta η → − η with k , η ∈ R k,\eta\in\mathbb{R} k , η ∈ R . We will show in this subsection that this symmetry forces the Bogoliubov β k \beta_k β k to be purely imaginary. Throughout this subsection, we will assume k , η ∈ R k,\eta\in\mathbb{R} k , η ∈ R .
We start with representing the time-reversal symmetry on the solution space S \mathcal{S} S (defined above eq. (14) ), which has two bases { χ k , 1 ( η ) , χ k , 1 ∗ ( η ) } \{\chi_{k,1}(\eta),\chi_{k,1}^{*}(\eta)\} { χ k , 1 ( η ) , χ k , 1 ∗ ( η )} and { χ k , 2 ( η ) , χ k , 2 ∗ ( η ) } \{\chi_{k,2}(\eta),\chi_{k,2}^{*}(\eta)\} { χ k , 2 ( η ) , χ k , 2 ∗ ( η )} defined in eq. (13) . As χ k , 1 ( η ) ∈ S \chi_{k,1}(\eta) \in\mathcal{S} χ k , 1 ( η ) ∈ S , under η → − η \eta\to-\eta η → − η the mode equation (29) for χ k , 1 ( η ) \chi_{k,1}(\eta) χ k , 1 ( η ) goes to
[ ∂ η 2 + ω k 2 ( η ) ] χ k , 1 ( − η ) = 0 , (75) \left[\partial_{\eta}^{2}+\omega_{k}^{2}(\eta)\right]\chi_{k,1}(-\eta)=0, \tag*{(75)} [ ∂ η 2 + ω k 2 ( η ) ] χ k , 1 ( − η ) = 0 , (75)
where we have used ω k ( η ) = ω k ( − η ) \omega_k(\eta)=\omega_k(-\eta) ω k ( η ) = ω k ( − η ) , which means χ k , 1 ( − η ) ∈ S \chi_{k,1}(-\eta) \in\mathcal{S} χ k , 1 ( − η ) ∈ S . We then expand χ k , 1 ( − η ) \chi_{k,1}(-\eta) χ k , 1 ( − η ) over the basis { χ k , 2 ( η ) , χ k , 2 ∗ ( η ) } \{\chi_{k,2}(\eta),\chi_{k,2}^{*}(\eta)\} { χ k , 2 ( η ) , χ k , 2 ∗ ( η )} as
χ k , 1 ( − η ) = c χ k , 2 ( η ) + d χ k , 2 ∗ ( η ) (76) \chi_{k,1}(-\eta)=c\chi_{k,2}(\eta)+d\chi_{k,2}^{*}(\eta) \tag*{(76)} χ k , 1 ( − η ) = c χ k , 2 ( η ) + d χ k , 2 ∗ ( η ) (76)
with c c c and d d d constants. These constants can be obtained by investigating the asymptotic behavior of χ k , 1 ( − η ) \chi_{k,1}(-\eta) χ k , 1 ( − η ) . Note as η \eta η approaches the asymptotic future η + ∞ \eta_{+\infty} η + ∞ , − η -\eta − η approaches the far past η − ∞ \eta_{-\infty} η − ∞ , meaning that
lim η → η + ∞ χ k , 1 ( − η ) = lim − η → η − ∞ e − i ∫ 0 − η d η ˉ ω k ( η ˉ ) 2 ω k ( − η ) = lim η → η + ∞ e i ∫ 0 η d η ˉ ω k ( η ˉ ) 2 ω k ( η ) , (77) \lim_{\eta\to\eta_{+\infty}}\chi_{k,1}(-\eta)=\lim_{-\eta\to\eta_{-\infty}}\frac{e^{-i\int_{0}^{-\eta}d\bar{\eta}\omega_k(\bar{\eta})}}{\sqrt{2\omega_k(-\eta)}}=\lim_{\eta\to\eta_{+\infty}}\frac{e^{i\int_{0}^{\eta}d\bar{\eta}\omega_k(\bar{\eta})}}{\sqrt{2\omega_k(\eta)}}, \tag*{(77)} η → η + ∞ lim χ k , 1 ( − η ) = − η → η − ∞ lim 2 ω k ( − η ) e − i ∫ 0 − η d η ˉ ω k ( η ˉ ) = η → η + ∞ lim 2 ω k ( η ) e i ∫ 0 η d η ˉ ω k ( η ˉ ) , (77)
where we have used ω k ( η ) = ω k ( − η ) \omega_k(\eta)=\omega_k(-\eta) ω k ( η ) = ω k ( − η ) again and η + ∞ = − η − ∞ \eta_{+\infty}=-\eta_{-\infty} η + ∞ = − η − ∞ . Note the right hand side of the above equation is identical to the boundary condition of χ k , 2 ∗ ( η ) \chi_{k,2}^{*}(\eta) χ k , 2 ∗ ( η ) , giving
lim η → η + ∞ χ k , 1 ( − η ) = lim η → η + ∞ χ k , 2 ∗ ( η ) , (78) \lim_{\eta\to\eta_{+\infty}}\chi_{k,1}(-\eta)=\lim_{\eta\to\eta_{+\infty}}\chi_{k,2}^{*}(\eta), \tag*{(78)} η → η + ∞ lim χ k , 1 ( − η ) = η → η + ∞ lim χ k , 2 ∗ ( η ) , (78)
then from the expansion (76) we obtain c = 0 c=0 c = 0 and d = 1 d=1 d = 1 , yielding
χ k , 1 ( − η ) = χ k , 2 ∗ ( η ) . (79) \chi_{k,1}(-\eta)=\chi_{k,2}^{*}(\eta). \tag*{(79)} χ k , 1 ( − η ) = χ k , 2 ∗ ( η ) . (79)
This leads to
χ k , 2 ( − η ) = χ k , 1 ∗ ( η ) . (80) \chi_{k,2}(-\eta)=\chi_{k,1}^{*}(\eta). \tag*{(80)} χ k , 2 ( − η ) = χ k , 1 ∗ ( η ) . (80)
Next, the Wronskian projection (16) implies
β k = i [ χ k , 1 ( η ) ∂ η χ k , 2 ( η ) − χ k , 2 ( η ) ∂ η χ k , 1 ( η ) ] (81) \beta_k=i\left[\chi_{k,1}(\eta)\partial_\eta\chi_{k,2}(\eta)-\chi_{k,2}(\eta)\partial_\eta\chi_{k,1}(\eta)\right] \tag*{(81)} β k = i [ χ k , 1 ( η ) ∂ η χ k , 2 ( η ) − χ k , 2 ( η ) ∂ η χ k , 1 ( η ) ] (81)
= i [ χ k , 1 ( − η ) ∂ − η χ k , 2 ( − η ) − χ k , 2 ( − η ) ∂ − η χ k , 1 ( − η ) ] , (82) =i\left[\chi_{k,1}(-\eta)\partial_{-\eta}\chi_{k,2}(-\eta)-\chi_{k,2}(-\eta)\partial_{-\eta}\chi_{k,1}(-\eta)\right], \tag*{(82)} = i [ χ k , 1 ( − η ) ∂ − η χ k , 2 ( − η ) − χ k , 2 ( − η ) ∂ − η χ k , 1 ( − η ) ] , (82)
where in the second line we have changed the time variable to − η -\eta − η since β k \beta_k β k is time-independent. On one hand, substituting eqs. (79) and (80) into eq. (82) yields
β k = i [ χ k , 1 ∗ ( η ) ∂ η χ k , 2 ∗ ( η ) − χ k , 2 ∗ ( η ) ∂ η χ k , 1 ∗ ( η ) ] . (83) \beta_k=i\left[\chi_{k,1}^{*}(\eta)\partial_\eta\chi_{k,2}^{*}(\eta)-\chi_{k,2}^{*}(\eta)\partial_\eta\chi_{k,1}^{*}(\eta)\right]. \tag*{(83)} β k = i [ χ k , 1 ∗ ( η ) ∂ η χ k , 2 ∗ ( η ) − χ k , 2 ∗ ( η ) ∂ η χ k , 1 ∗ ( η ) ] . (83)
On the other hand, conjugating eq. (81) gives
β k ∗ = − i [ χ k , 1 ∗ ( η ) ∂ η χ k , 2 ∗ ( η ) − χ k , 2 ∗ ( η ) ∂ η χ k , 1 ∗ ( η ) ] . (84) \beta_k^{*}=-i\left[\chi_{k,1}^{*}(\eta)\partial_\eta\chi_{k,2}^{*}(\eta)-\chi_{k,2}^{*}(\eta)\partial_\eta\chi_{k,1}^{*}(\eta)\right]. \tag*{(84)} β k ∗ = − i [ χ k , 1 ∗ ( η ) ∂ η χ k , 2 ∗ ( η ) − χ k , 2 ∗ ( η ) ∂ η χ k , 1 ∗ ( η ) ] . (84)
Hence,
β k ∗ = − β k , (85) \beta_k^{*}=-\beta_k, \tag*{(85)} β k ∗ = − β k , (85)
which means that the Bogoliubov β k \beta_k β k must be purely imaginary as a result of the time-reversal symmetry.
The first Stokes constant
In section 3.1 we noted that all the Stokes constants are expressible in terms of the first Stokes constant function S 1 ( k ) S_1(k) S 1 ( k ) with rotated argument and certain phase factors. As may be seen from the complexified version of the amplitude ODE (24) or its integrated version
α k ( z ) = α k ( z 0 ) + ∫ z 0 z d z ′ ω k ′ ( z ′ ) 2 ω k ( z ′ ) e + 2 i θ ( k , z ′ ) β k ( z ′ ) (86) \alpha_k(z)=\alpha_k(z_0)+\int_{z_0}^{z} dz'\,\frac{\omega'_k(z')}{2\omega_k(z')}e^{+2i\theta(k,z')}\beta_k(z') \tag*{(86)} α k ( z ) = α k ( z 0 ) + ∫ z 0 z d z ′ 2 ω k ( z ′ ) ω k ′ ( z ′ ) e + 2 i θ ( k , z ′ ) β k ( z ′ ) (86)
β k ( z ) = β k ( z 0 ) + ∫ z 0 z d z ′ ω k ′ ( z ′ ) 2 ω k ( z ′ ) e − 2 i θ ( k , z ′ ) α k ( z ′ ) (87) \beta_k(z)=\beta_k(z_0)+\int_{z_0}^{z} dz'\,\frac{\omega'_k(z')}{2\omega_k(z')}e^{-2i\theta(k,z')}\alpha_k(z') \tag*{(87)} β k ( z ) = β k ( z 0 ) + ∫ z 0 z d z ′ 2 ω k ( z ′ ) ω k ′ ( z ′ ) e − 2 i θ ( k , z ′ ) α k ( z ′ ) (87)
an explicit computation of the WKB amplitudes, thereby the Stokes constants, requires integrals of oscillatory exponentials along contours in the asymptotic Stokes sectors. To mitigate this obstacle, one may naively consider evaluating the asymptotic expansions of eqs. (86) and (87) instead, which seems to simplify the computation of the Stokes constants. However, ordinary asymptotic expansions are not accurate enough to capture the Stokes constants. To see this, consider the exact mode decomposition over the fictitious WKB basis χ k ( z ) = α k ( z ) f − ( k , z ) + β k ( z ) f + ( k , z ) \chi_k(z)=\alpha_k(z)f_-(k,z)+\beta_k(z)f_+(k,z) χ k ( z ) = α k ( z ) f − ( k , z ) + β k ( z ) f + ( k , z ) . In the first Stokes sector S 1 \mathcal{S}_1 S 1 and in our branch convention, f − f_- f − and f + f_+ f + are exponentially amplified and suppressed by ∣ z ∣ |z| ∣ z ∣ , respectively. This means the first Stokes constant S 1 ( k ) S_1(k) S 1 ( k ) , an O ( 1 ) O(1) O ( 1 ) shift in β k ( z ) \beta_k(z) β k ( z ) , is exponentially suppressed in the mode decomposition and is invisible to all algebraic order asymptotic contributions. Therefore, to compute S 1 ( k ) S_1(k) S 1 ( k ) by evaluating the asymptotic behavior of eqs. (86) and (87) , one must go beyond ordinary asymptotic expansions to take the exponentially suppressed contributions into account. We therefore consider the exponentially improved asymptotic expansion of the WKB amplitudes as ∣ z ∣ → ∞ |z|\to\infty ∣ z ∣ → ∞ , which can be schematically expressed as
V k ( z ) ∼ ˙ constant + power suppressed + exponentially suppressed (88) V_k(z)\dot{\sim}\text{constant}+\text{power suppressed}+\text{exponentially suppressed} \tag*{(88)} V k ( z ) ∼ ˙ constant + power suppressed + exponentially suppressed (88)
where ∼ ˙ \dot{\sim} ∼ ˙ denotes exponentially improved asymptotic expansion (which we will often abbreviate as “e.a.e.”).16
After plugging the exponentially improved asymptotic expansions of the WKB amplitudes into eqs. (86) and (87) , we encounter integrals of the form
E p ( z ) ≡ z p − 1 ∫ z ∞ d t e − t t p , z ≠ 0. (89) E_p(z)\equiv z^{p-1}\int_z^\infty dt\,\frac{e^{-t}}{t^p},\qquad z\ne0. \tag*{(89)} E p ( z ) ≡ z p − 1 ∫ z ∞ d t t p e − t , z = 0. (89)
These are called generalized exponential integrals and have known exponentially improved asymptotic expansions given by
E p ( z ) ∼ ˙ e − z z ∑ s = 0 ∞ ( − 1 ) s ( p ) s z s + Θ R ( arg z − π ) 2 π i e − i p π Γ ( p ) z p − 1 , − 3 π 2 < arg z < 7 π 2 (90) E_p(z)\dot{\sim}\frac{e^{-z}}{z}\sum_{s=0}^{\infty}\frac{(-1)^s(p)_s}{z^s}+\Theta_{\mathbb{R}}(\arg z-\pi)2\pi i\frac{e^{-ip\pi}}{\Gamma(p)}z^{p-1},\qquad-\frac{3\pi}{2}<\arg z<\frac{7\pi}{2} \tag*{(90)} E p ( z ) ∼ ˙ z e − z s = 0 ∑ ∞ z s ( − 1 ) s ( p ) s + Θ R ( arg z − π ) 2 π i Γ ( p ) e − i p π z p − 1 , − 2 3 π < arg z < 2 7 π (90)
where Θ R ( x ) \Theta_{\mathrm{R}}(x) Θ R ( x ) denotes a step-function that jumps at x = 0 x = 0 x = 0 with Θ R ( 0 ) = 1 / 2 \Theta_{\mathrm{R}}(0) = 1/2 Θ R ( 0 ) = 1/2 (see appendix § 13 ). As seen in the above, these expansions undergo Stokes jumps, with known Stokes constants, as z z z is analytically continued across the negative real line at arg z = π \arg z = \pi arg z = π .17
Given the expansions (88) and (90) , we note that the coupled integrated equations (86) and (87) could be understood as an automorphism on V k \mathcal{V}_k V k — the functional space of exponentially improved asymptotic expansions of the WKB amplitudes, denoted as [ V k ( z ) ] e . a . e . [\mathcal{V}_k(z)]_{\mathrm{e.a.e.}} [ V k ( z ) ] e.a.e. . In particular, plugging [ V k ( z ) ] e . a . e . [\mathcal{V}_k(z)]_{\mathrm{e.a.e.}} [ V k ( z ) ] e.a.e. given by eq. (88) into the integrals on the right hand side of eqs. (86) and (87) and computing them term by term, we find both
∫ z 0 z d z ′ ω k ′ ( z ′ ) 2 ω k ( z ′ ) e + 2 i θ ( k , z ′ ) [ β k ( z ′ ) ] e . a . e . and ∫ z 0 z d z ′ ω k ′ ( z ′ ) 2 ω k ( z ′ ) e − 2 i θ ( k , z ′ ) [ α k ( z ′ ) ] e . a . e . (91) \int_{z_0}^{z} dz'\,\frac{\omega_k'(z')}{2\omega_k(z')}e^{+2i\theta(k,z')}[\beta_k(z')]_{\mathrm{e.a.e.}} \quad\text{and} \quad\int_{z_0}^{z} dz'\,\frac{\omega_k'(z')}{2\omega_k(z')}e^{-2i\theta(k,z')}[\alpha_k(z')]_{\mathrm{e.a.e.}} \tag*{(91)} ∫ z 0 z d z ′ 2 ω k ( z ′ ) ω k ′ ( z ′ ) e + 2 i θ ( k , z ′ ) [ β k ( z ′ ) ] e.a.e. and ∫ z 0 z d z ′ 2 ω k ( z ′ ) ω k ′ ( z ′ ) e − 2 i θ ( k , z ′ ) [ α k ( z ′ ) ] e.a.e. (91)
have exponentially improved asymptotic expansions belonging to V k \mathcal{V}_k V k . Therefore, we may view
∫ z 0 z d z ′ ω k ′ ( z ′ ) 2 ω k ( z ′ ) e ± 2 i θ ( k , z ′ ) (92) \int_{z_0}^{z} dz'\,\frac{\omega_k'(z')}{2\omega_k(z')}e^{\pm2i\theta(k,z')} \tag*{(92)} ∫ z 0 z d z ′ 2 ω k ( z ′ ) ω k ′ ( z ′ ) e ± 2 i θ ( k , z ′ ) (92)
acting on [ β k ( z ′ ) ] e . a . e . [\beta_k(z')]_{\mathrm{e.a.e.}} [ β k ( z ′ ) ] e.a.e. and [ α k ( z ′ ) ] e . a . e . [\alpha_k(z')]_{\mathrm{e.a.e.}} [ α k ( z ′ ) ] e.a.e. as an automorphism on V k \mathcal{V}_k V k . We will see that this automorphism yields a set of recursion relations on the expansion coefficients of [ α k ( z ) ] e . a . e . [\alpha_k(z)]_{\mathrm{e.a.e.}} [ α k ( z ) ] e.a.e. and [ β k ( z ) ] e . a . e . [\beta_k(z)]_{\mathrm{e.a.e.}} [ β k ( z ) ] e.a.e. , whose solution leads to a power series expression of the first Stokes constant.
The rest of this section is organized as follows. In section 4.1, we decompose the WKB phase θ ( k , z ) \theta(k,z) θ ( k , z ) in the first Stokes sector into three components: the leading term z n / 2 + 1 / ( n / 2 + 1 ) z^{n/2+1}/(n/2+1) z n /2 + 1 / ( n /2 + 1 ) , a power series of 1 / z 1/z 1/ z , and a constant factor non-analytic in k k k . Then, in section 4.2, we show that this decomposition allows us to compute the integrals in eq. (91) using the expression (90) for the generalized exponential integrals. This computation leads to a detailed representation of the automorphism on V k \mathcal{V}_k V k . In section 4.3, we show how to translate this automorphism to a set of recursion relations of the expansion coefficients of the WKB amplitudes, which is solved in section 4.4, yielding a closed form expression for the first Stokes constant up to quadratic order.
The above method follows [16 , 18 ] up to the automorphism perspective and the choice of the WKB basis functions. The WKB basis functions in [16 , 18 ] were chosen to capture only the exponential asymptotics implied by the dominant essential singularity at ∣ z ∣ → ∞ |z| \to\infty ∣ z ∣ → ∞ . In our formalism, we have instead chosen a WKB basis more familiar to physicists and suited to the rotational symmetry arguments of section 3. As we will show in section 4.1, this choice primarily differs from [16 , 18 ] by a constant factor non-analytic in k 2 k^2 k 2 — the non-analyticity arising due to contributions to the phase d z dz d z integral between 0 0 0 and ∣ z ∣ ≤ k 2 / n |z| \leq k^{2/n} ∣ z ∣ ≤ k 2/ n . Up to this factor, our WKB basis (and that in [16 , 18 ] ) is entire in k 2 k^2 k 2 .18
Non-analyticity isolation
In this subsection, we note that the WKB basis functions can be written as a product of a z z z -independent factor non-analytic in k 2 k^2 k 2 and a z z z -dependent entire function of k 2 k^2 k 2 . The latter captures the exponential asymptotics implied by the irregular singularity at ∣ z ∣ → ∞ |z| \to\infty ∣ z ∣ → ∞ . Consider the WKB phase
θ ( k , z ) ≡ ∫ 0 z d z ′ ω k ( z ′ ) , ω k ( z ′ ) = k 2 + z ′ n (93) \theta(k,z) \equiv\int_{0}^{z} dz'\,\omega_k(z'), \qquad\omega_k(z') = \sqrt{k^2+z'^{n}} \tag*{(93)} θ ( k , z ) ≡ ∫ 0 z d z ′ ω k ( z ′ ) , ω k ( z ′ ) = k 2 + z ′ n (93)
where the non-analyticity in k 2 k^2 k 2 arises from integration between 0 0 0 and ∣ z ∣ ≤ k 2 / n |z| \le k^{2/n} ∣ z ∣ ≤ k 2/ n . Consequently, we may separate the two contributions as
θ ( k , z ) = ( analytic in k ) + ( non-analytic in k ) . (94) \theta(k,z) = (\text{analytic in } k) + (\text{non-analytic in } k). \tag*{(94)} θ ( k , z ) = ( analytic in k ) + ( non-analytic in k ) . (94)
by splitting the integral at ψ 1 \psi_1 ψ 1 (since we are interested in the first Stokes sector), that is
θ ( k , z ) = ∫ 0 ψ 1 d z ′ ω k ( z ′ ) + ∫ ψ 1 z d z ′ ω k ( z ′ ) , z ∈ S 1 . (95) \theta(k,z) = \int_{0}^{\psi_1} dz'\,\omega_k(z') + \int_{\psi_1}^{z} dz'\,\omega_k(z'), \qquad z \in\mathcal{S}_1. \tag*{(95)} θ ( k , z ) = ∫ 0 ψ 1 d z ′ ω k ( z ′ ) + ∫ ψ 1 z d z ′ ω k ( z ′ ) , z ∈ S 1 . (95)
The first integral is the phase factor Δ 1 ( k ) \Delta_1(k) Δ 1 ( k ) and is non-analytic in k 2 k^2 k 2 . The second integral can be evaluated term by term using the Taylor series representations (A.2) and (A.8). Evaluating the resulting primitive of this Taylor series at ψ 1 \psi_1 ψ 1 then yields another term non-analytic in k 2 k^2 k 2 . Combining these two terms we obtain for n ≥ 4 n \ge4 n ≥ 4 in the first Stokes sector
θ ( k , z ) = z n / 2 + 1 n / 2 + 1 − A ( k , z ) ⏟ analytic in k 2 + θ ˉ ( k ) ⏟ non-analytic in k , k ∈ R ≥ 0 , z ∈ S 1 (96) \theta(k,z) = \frac{z^{n/2+1}}{n/2+1} - \underbrace{A(k,z)}_{\text{analytic in }k^2} + \underbrace{\bar{\theta}(k)}_{\text{non-analytic in }k}, \qquad k \in\mathbb{R}_{\ge0},\ z \in\mathcal{S}_1 \tag*{(96)} θ ( k , z ) = n /2 + 1 z n /2 + 1 − analytic in k 2 A ( k , z ) + non-analytic in k θ ˉ ( k ) , k ∈ R ≥ 0 , z ∈ S 1 (96)
where
θ ˉ ( k ) ≡ Δ 1 ( k ) + Ψ 1 ( k ) (97) \bar{\theta}(k) \equiv\Delta_1(k) + \Psi_1(k) \tag*{(97)} θ ˉ ( k ) ≡ Δ 1 ( k ) + Ψ 1 ( k ) (97)
denotes the term non-analytic in k k k and where explicit expressions for A ( k , z ) A(k,z) A ( k , z ) and Ψ 1 ( k ) \Psi_1(k) Ψ 1 ( k ) are given by Eqs. (B.5) and (B.6), respectively, as
A ( k , z ) = ∑ m = 1 ∞ ( 1 / 2 m ) k 2 m 2 m N − ( N + 1 ) 1 z 2 m N − ( N + 1 ) (98) A(k,z) = \sum_{m=1}^{\infty} \binom{1/2}{m} \frac{k^{2m}}{2mN-(N+1)} \frac{1}{z^{2mN-(N+1)}} \tag*{(98)} A ( k , z ) = m = 1 ∑ ∞ ( m 1/2 ) 2 m N − ( N + 1 ) k 2 m z 2 m N − ( N + 1 ) 1 (98)
Ψ 1 ( k ) = ∑ m = 0 ∞ ( 1 / 2 m ) 1 m n − ( n / 2 + 1 ) k 2 m ψ 1 m n − ( n / 2 + 1 ) , (99) \Psi_1(k) = \sum_{m=0}^{\infty} \binom{1/2}{m} \frac{1}{mn-(n/2+1)} \frac{k^{2m}}{\psi_1^{mn-(n/2+1)}}, \tag*{(99)} Ψ 1 ( k ) = m = 0 ∑ ∞ ( m 1/2 ) mn − ( n /2 + 1 ) 1 ψ 1 mn − ( n /2 + 1 ) k 2 m , (99)
where this expansion breaks down for n = 2 n=2 n = 2 (N = n / 2 = 1 N=n/2=1 N = n /2 = 1 ) for example because of the denominator m n − ( n / 2 + 1 ) mn-(n/2+1) mn − ( n /2 + 1 ) appearing in the expression for Ψ 1 \Psi_1 Ψ 1 .19
In appendix E, we show that the non-analytic residue θ ˉ ( k ) \bar{\theta}(k) θ ˉ ( k ) is proportional to the phase factor Δ 1 ( k ) \Delta_1(k) Δ 1 ( k ) :
θ ˉ ( k ) = e − i π / n cos ( π / n ) Δ 1 ( k ) . (100) \bar{\theta}(k) = \frac{e^{-i\pi/n}}{\cos(\pi/n)}\Delta_1(k). \tag*{(100)} θ ˉ ( k ) = cos ( π / n ) e − iπ / n Δ 1 ( k ) . (100)
We will also see in section 5 that this cancels the non-analytic phase factors arising from branch cuts, making the Bogoliubov β k \beta_k β k analytic in k 2 k^2 k 2 in the soft limit.
that the Stokes constants will be a product of a non-analytic factor and an entire function of k 2 k^2 k 2 — which we derive in (119) .
Automorphism
Here we show in detail how using the WKB amplitude expansion (88) in eqs. (86) and (87) leads to an automorphism on V k \mathcal{V}_k V k , the functional space of the exponentially improved asymptotic expansion of the WKB amplitudes. The main result is eq. (115) , the explicit form of V k \mathcal{V}_k V k .
We start with simplifying eqs. (86) and (87) by taking z 0 → ∞ z_0 \to\infty z 0 → ∞ on the real axis and the boundary condition20
lim z → z 0 V k ( z ) = lim z → z 0 ( α k ( z ) β k ( z ) ) = ( 1 0 ) . (101) \lim_{z \to z_0} V_k(z) = \lim_{z \to z_0} \begin{pmatrix} \alpha_k(z) \\ \beta_k(z) \end{pmatrix} = \begin{pmatrix} 1 \\ 0 \end{pmatrix}. \tag*{(101)} z → z 0 lim V k ( z ) = z → z 0 lim ( α k ( z ) β k ( z ) ) = ( 1 0 ) . (101)
Using the decomposition (96) for the WKB phase in the first Stokes sector, we obtain
α k ( z ) = 1 + ∫ ∞ z d w e ζ ( w ) B − ( w ) e 2 i θ ˉ ( k ) β k ( w ) , (102) \alpha_k(z) = 1 + \int_{\infty}^{z} dw\, e^{\zeta(w)} B_-(w)e^{2i\bar{\theta}(k)}\beta_k(w), \tag*{(102)} α k ( z ) = 1 + ∫ ∞ z d w e ζ ( w ) B − ( w ) e 2 i θ ˉ ( k ) β k ( w ) , (102)
β k ( z ) = ∫ ∞ z d w e − ζ ( w ) B + ( w ) e − 2 i θ ˉ ( k ) α k ( w ) . (103) \beta_k(z) = \int_{\infty}^{z} dw\, e^{-\zeta(w)} B_+(w)e^{-2i\bar{\theta}(k)}\alpha_k(w). \tag*{(103)} β k ( z ) = ∫ ∞ z d w e − ζ ( w ) B + ( w ) e − 2 i θ ˉ ( k ) α k ( w ) . (103)
where ζ ( w ) ≡ 2 i w N + 1 / ( N + 1 ) \zeta(w) \equiv2iw^{N+1}/(N+1) ζ ( w ) ≡ 2 i w N + 1 / ( N + 1 ) and
B ± ( z ) ≡ ω k ′ ( z ) 2 ω k ( z ) e ± 2 i A ( k , z ) = ∑ m = 1 ∞ B m ( ± ) z − m , z ∈ S 1 (104) B_{\pm}(z) \equiv\frac{\omega_k'(z)}{2\omega_k(z)}e^{\pm2iA(k,z)} = \sum_{m=1}^{\infty}B_m^{(\pm)}z^{-m}, \qquad z \in\mathcal{S}_1 \tag*{(104)} B ± ( z ) ≡ 2 ω k ( z ) ω k ′ ( z ) e ± 2 i A ( k , z ) = m = 1 ∑ ∞ B m ( ± ) z − m , z ∈ S 1 (104)
with B m ( ± ) B_m^{(\pm)} B m ( ± ) denoting expansion coefficients of B ± ( z ) B_{\pm}(z) B ± ( z ) .
To make the expansion (88) concrete, we define the functional space
P ( z ) ≡ { ∑ m = 1 ∞ a m z − m ∣ a m ∈ C } (105) \mathcal{P}(z) \equiv\left\{\sum_{m=1}^{\infty}a_m z^{-m}\mid a_m \in\mathbb{C}\right\} \tag*{(105)} P ( z ) ≡ { m = 1 ∑ ∞ a m z − m ∣ a m ∈ C } (105)
allowing us to classify B ± ( z ) ∈ P ( z ) B_{\pm}(z) \in\mathcal{P}(z) B ± ( z ) ∈ P ( z ) . Also note in the asymptotic regions, the exponentially improved asymptotic expansions of the WKB amplitudes [ α k ( z ) ] e . a . e . [\alpha_k(z)]_{\mathrm{e.a.e.}} [ α k ( z ) ] e.a.e. and [ β k ( z ) ] e . a . e . [\beta_k(z)]_{\mathrm{e.a.e.}} [ β k ( z ) ] e.a.e. must contain z − p z^{-p} z − p terms because of the structure of eqs. (102) –(104) . Plugging such terms into eqs. (102) and (103) yields terms of the form
I p ± ( z ) ≡ ∫ ∞ z d w e ± ζ ( w ) w p . (106) I_p^{\pm}(z) \equiv\int_{\infty}^{z} dw \frac{e^{\pm\zeta(w)}}{w^p}. \tag*{(106)} I p ± ( z ) ≡ ∫ ∞ z d w w p e ± ζ ( w ) . (106)
A change of variable w → ± ζ = ± 2 i w N + 1 / ( N + 1 ) w \to\pm\zeta= \pm2iw^{N+1}/(N+1) w → ± ζ = ± 2 i w N + 1 / ( N + 1 ) gives
I p ± ( z ) ∝ ζ 1 − N + p N + 1 E N + p N + 1 ( ∓ ζ ) . (107) I_p^{\pm}(z) \propto\zeta^{1-\frac{N+p}{N+1}}E_{\frac{N+p}{N+1}}(\mp\zeta). \tag*{(107)} I p ± ( z ) ∝ ζ 1 − N + 1 N + p E N + 1 N + p ( ∓ ζ ) . (107)
Since we are interested in the first Stokes sector where arg ζ ∈ ( π / 2 , 3 π / 2 ) \arg\zeta\in(\pi/2,3\pi/2) arg ζ ∈ ( π /2 , 3 π /2 ) as arg w ∈ ( 0 , π / ( N + 1 ) ) \arg w \in(0,\pi/(N+1)) arg w ∈ ( 0 , π / ( N + 1 )) , the asymptotic expansion of I p − ( z ) I_p^-(z) I p − ( z ) sees a Stokes jump as arg ζ \arg\zeta arg ζ crosses π \pi π , or equivalently, as arg w \arg w arg w crosses π / ( 2 N + 2 ) \pi/(2N+2) π / ( 2 N + 2 ) . Consequently, as may be seen from eq. (90) , the exponentially improved asymptotic expansions of these integrals in S 1 \mathcal{S}_1 S 1 take the form
[ ∫ ∞ z d w e ζ ( w ) w p ] e . a . e . ∈ e ζ ( z ) P ( z ) (108) \left[\int_{\infty}^{z} \mathrm{d}w \frac{e^{\zeta(w)}}{w^p}\right]_{\mathrm{e.a.e.}} \in e^{\zeta(z)}\mathcal{P}(z) \tag*{(108)} [ ∫ ∞ z d w w p e ζ ( w ) ] e.a.e. ∈ e ζ ( z ) P ( z ) (108)
[ ∫ ∞ z d w e − ζ ( w ) w p ] e . a . e . ∈ e − ζ ( z ) P ( z ) + Θ ( z ) C (109) \left[\int_{\infty}^{z} \mathrm{d}w \frac{e^{-\zeta(w)}}{w^p}\right]_{\mathrm{e.a.e.}} \in e^{-\zeta(z)}\mathcal{P}(z)+\Theta(z)\mathbb{C} \tag*{(109)} [ ∫ ∞ z d w w p e − ζ ( w ) ] e.a.e. ∈ e − ζ ( z ) P ( z ) + Θ ( z ) C (109)
where Θ ( z ) \Theta(z) Θ ( z ) is a step-function that jumps at the bisector of the first Stokes sector
Θ ( z ) ≡ { 0 0 ≤ arg z < π 2 ( N + 1 ) 1 / 2 arg z = π 2 ( N + 1 ) 1 π 2 ( N + 1 ) < arg z ≤ π N + 1 (110) \Theta(z)\equiv
\begin{cases}
0 & 0\leq\arg z<\frac{\pi}{2(N+1)} \\
1/2 & \arg z=\frac{\pi}{2(N+1)} \\
1 & \frac{\pi}{2(N+1)}<\arg z\leq\frac{\pi}{N+1}
\end{cases} \tag*{(110)} Θ ( z ) ≡ ⎩ ⎨ ⎧ 0 1/2 1 0 ≤ arg z < 2 ( N + 1 ) π arg z = 2 ( N + 1 ) π 2 ( N + 1 ) π < arg z ≤ N + 1 π (110)
Now since α k ( z ) \alpha_k(z) α k ( z ) is the coefficient of the dominant mode f − f_- f − in S 1 \mathcal{S}_1 S 1 , which doesn’t see a non-exponentially-suppressed Θ ( z ) \Theta(z) Θ ( z ) and satisfies the boundary condition α k ( z 0 ) = 1 \alpha_k(z_0)=1 α k ( z 0 ) = 1 , we have the ordinary asymptotic expansion (without exponential improvements) of α k ( z ) \alpha_k(z) α k ( z ) having the form 1 + P ( z ) 1+\mathcal{P}(z) 1 + P ( z ) . Plugging this into eqs. (102) and (103) and using the generalization of eq. (108) and eq. (109) where w − p w^{-p} w − p is replaced by an element of P ( w ) \mathcal{P}(w) P ( w ) , that is
∫ ∞ z d w e ζ ( w ) P ( w ) ∼ ˙ e ζ ( z ) P ( z ) (111) \int_{\infty}^{z}\mathrm{d}w e^{\zeta(w)}\mathcal{P}(w)\dot{\sim}e^{\zeta(z)}\mathcal{P}(z) \tag*{(111)} ∫ ∞ z d w e ζ ( w ) P ( w ) ∼ ˙ e ζ ( z ) P ( z ) (111)
∫ ∞ z d w e − ζ ( w ) P ( w ) ∼ ˙ e − ζ ( z ) P ( z ) + Θ ( z ) C (112) \int_{\infty}^{z}\mathrm{d}w e^{-\zeta(w)}\mathcal{P}(w)\dot{\sim}e^{-\zeta(z)}\mathcal{P}(z)+\Theta(z)\mathbb{C} \tag*{(112)} ∫ ∞ z d w e − ζ ( w ) P ( w ) ∼ ˙ e − ζ ( z ) P ( z ) + Θ ( z ) C (112)
we note that the exponentially improved asymptotic expansions of the WKB amplitudes should take the form (for a more rigorous justification of the following, see appendix H)
[ α k ( z ) ] e . a . e . ∈ 1 + P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) (113) [\alpha_k(z)]_{\mathrm{e.a.e.}}\in1+\mathcal{P}(z)+e^{\zeta(z)}\Theta(z)\mathcal{P}(z) \tag*{(113)} [ α k ( z ) ] e.a.e. ∈ 1 + P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) (113)
[ β k ( z ) ] e . a . e . ∈ e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] (114) [\beta_k(z)]_{\mathrm{e.a.e.}}\in e^{-\zeta(z)}\mathcal{P}(z)+\Theta(z)[\mathcal{P}(z)+\mathbb{C}] \tag*{(114)} [ β k ( z ) ] e.a.e. ∈ e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] (114)
This is equivalent to
V k = ( 1 + P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] ) (115) \mathcal{V}_k=
\begin{pmatrix}
1+\mathcal{P}(z)+e^{\zeta(z)}\Theta(z)\mathcal{P}(z) \\
e^{-\zeta(z)}\mathcal{P}(z)+\Theta(z)[\mathcal{P}(z)+\mathbb{C}]
\end{pmatrix} \tag*{(115)} V k = ( 1 + P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] ) (115)
where recall that V k \mathcal{V}_k V k denotes functional space of the exponentially improved asymptotic expansions of the WKB amplitudes. Consequently, the integrals on the right hand side of eqs. (102) and (103) may be viewed as automorphisms on V k \mathcal{V}_k V k via series multiplications by B ± ( z ) B_{\pm}(z) B ± ( z ) , generalized exponential integrals (111) and (112) and ordinary integrals, which we will further detail in section 4.3.
For the purposes of the exact computations in the following section we define the power series
λ 1 , 2 ( z ) ≡ ∑ m = 1 ∞ λ m ( 1 , 2 ) z − m , μ 1 , 2 ( z ) ≡ ∑ m = 1 ∞ μ m ( 1 , 2 ) z − m (116) \lambda_{1,2}(z)\equiv\sum_{m=1}^{\infty}\lambda_m^{(1,2)}z^{-m},\qquad\mu_{1,2}(z)\equiv\sum_{m=1}^{\infty}\mu_m^{(1,2)}z^{-m} \tag*{(116)} λ 1 , 2 ( z ) ≡ m = 1 ∑ ∞ λ m ( 1 , 2 ) z − m , μ 1 , 2 ( z ) ≡ m = 1 ∑ ∞ μ m ( 1 , 2 ) z − m (116)
such that
[ α k ( z ) ] e . a . e . = 1 + λ 1 ( z ) + S 1 Θ ( z ) e ζ ( z ) μ 1 ( z ) (117) [\alpha_k(z)]_{\mathrm{e.a.e.}}=1+\lambda_1(z)+S_1\Theta(z)e^{\zeta(z)}\mu_1(z) \tag*{(117)} [ α k ( z ) ] e.a.e. = 1 + λ 1 ( z ) + S 1 Θ ( z ) e ζ ( z ) μ 1 ( z ) (117)
[ β k ( z ) ] e . a . e . = e − ζ ( z ) μ 2 ( z ) + S 1 Θ ( z ) + S 1 Θ ( z ) λ 2 ( z ) . (118) [\beta_k(z)]_{\mathrm{e.a.e.}}=e^{-\zeta(z)}\mu_2(z)+S_1\Theta(z)+S_1\Theta(z)\lambda_2(z). \tag*{(118)} [ β k ( z ) ] e.a.e. = e − ζ ( z ) μ 2 ( z ) + S 1 Θ ( z ) + S 1 Θ ( z ) λ 2 ( z ) . (118)
Here S 1 S_1 S 1 corresponds to the Stokes constant, and evidently λ 1 , 2 ( z ) \lambda_{1,2}(z) λ 1 , 2 ( z ) and μ 1 , 2 ( z ) \mu_{1,2}(z) μ 1 , 2 ( z ) belong to P ( z ) \mathcal{P}(z) P ( z ) .
Solvable technique
In this section, we work out the explicit expressions for the automorphism on V k \mathcal{V}_k V k given the series representation (117) and (118) for the exponentially improved asymptotic expansion of the WKB amplitudes. This will determine λ m ( 1 , 2 ) \lambda_m^{(1,2)} λ m ( 1 , 2 ) and μ m ( 1 , 2 ) \mu_m^{(1,2)} μ m ( 1 , 2 ) as functions of B m ( ± ) B_m^{(\pm)} B m ( ± ) , and an expression for S 1 ( k ) S_1(k) S 1 ( k ) as a function of these coefficients.
Rewriting eqs. (102) and (103) in terms of
β ˉ ( z ) ≡ e 2 i θ ˉ ( k ) β ( z ) , T 1 ≡ e 2 i θ ˉ ( k ) S 1 , (119) \bar{\beta}(z)\equiv e^{2i\bar{\theta}(k)}\beta(z),\qquad T_1\equiv e^{2i\bar{\theta}(k)}S_1, \tag*{(119)} β ˉ ( z ) ≡ e 2 i θ ˉ ( k ) β ( z ) , T 1 ≡ e 2 i θ ˉ ( k ) S 1 , (119)
we have
α ( z ) = 1 + ∫ ∞ z d w e ζ ( w ) B − ( w ) β ˉ ( w ) , (120) \alpha(z)=1+\int_{\infty}^{z}dw\,e^{\zeta(w)}B_-(w)\bar{\beta}(w), \tag*{(120)} α ( z ) = 1 + ∫ ∞ z d w e ζ ( w ) B − ( w ) β ˉ ( w ) , (120)
β ˉ ( z ) = ∫ ∞ z d w e − ζ ( w ) B + ( w ) α ( w ) . (121) \bar{\beta}(z)=\int_{\infty}^{z}dw\,e^{-\zeta(w)}B_+(w)\alpha(w). \tag*{(121)} β ˉ ( z ) = ∫ ∞ z d w e − ζ ( w ) B + ( w ) α ( w ) . (121)
Similarly, eqs. (117) and (118) can be rewritten as
[ α k ( z ) ] e . a . e . = 1 + λ 1 ( z ) + T 1 Θ ( z ) e ζ ( z ) μ ˉ 1 ( z ) (122) [\alpha_k(z)]_{\mathrm{e.a.e.}}=1+\lambda_1(z)+T_1\Theta(z)e^{\zeta(z)}\bar{\mu}_1(z) \tag*{(122)} [ α k ( z ) ] e.a.e. = 1 + λ 1 ( z ) + T 1 Θ ( z ) e ζ ( z ) μ ˉ 1 ( z ) (122)
[ β ˉ k ( z ) ] e . a . e . = e − ζ ( z ) μ ˉ 2 ( z ) + T 1 Θ ( z ) + T 1 Θ ( z ) λ 2 ( z ) (123) [\bar{\beta}_k(z)]_{\mathrm{e.a.e.}}=e^{-\zeta(z)}\bar{\mu}_2(z)+T_1\Theta(z)+T_1\Theta(z)\lambda_2(z) \tag*{(123)} [ β ˉ k ( z ) ] e.a.e. = e − ζ ( z ) μ ˉ 2 ( z ) + T 1 Θ ( z ) + T 1 Θ ( z ) λ 2 ( z ) (123)
where
μ ˉ 1 ( z ) ≡ e − 2 i θ ˉ ( k ) μ 1 ( z ) , μ ˉ 2 ( z ) ≡ e 2 i θ ˉ ( k ) μ 2 ( z ) . (124) \bar{\mu}_1(z)\equiv e^{-2i\bar{\theta}(k)}\mu_1(z),\qquad\bar{\mu}_2(z)\equiv e^{2i\bar{\theta}(k)}\mu_2(z). \tag*{(124)} μ ˉ 1 ( z ) ≡ e − 2 i θ ˉ ( k ) μ 1 ( z ) , μ ˉ 2 ( z ) ≡ e 2 i θ ˉ ( k ) μ 2 ( z ) . (124)
To carry out the integrals on the right hand side, we use eq. (90) which with the relevant change of variables implies in the first Stokes sector (see appendix F)
∫ ∞ z d w e ζ ( w ) w p ∼ ˙ e ζ ( z ) ∑ s = 0 ∞ E ˉ s , p z ( N + 1 ) s + N + p (125) \int_{\infty}^{z}dw\,\frac{e^{\zeta(w)}}{w^p}\mathrel{\dot{\sim}}e^{\zeta(z)}\sum_{s=0}^{\infty}\frac{\bar{E}_{s,p}}{z^{(N+1)s+N+p}} \tag*{(125)} ∫ ∞ z d w w p e ζ ( w ) ∼ ˙ e ζ ( z ) s = 0 ∑ ∞ z ( N + 1 ) s + N + p E ˉ s , p (125)
∫ ∞ z d w e − ζ ( w ) w p ∼ ˙ e − ζ ( z ) ∑ s = 0 ∞ ( − 1 ) s + 1 E ˉ s , p z ( N + 1 ) s + N + p − Θ ( z ) S ˉ N + p N + 1 (126) \int_{\infty}^{z}dw\,\frac{e^{-\zeta(w)}}{w^p}\mathrel{\dot{\sim}}e^{-\zeta(z)}\sum_{s=0}^{\infty}\frac{(-1)^{s+1}\bar{E}_{s,p}}{z^{(N+1)s+N+p}}-\Theta(z)\bar{S}_{\frac{N+p}{N+1}} \tag*{(126)} ∫ ∞ z d w w p e − ζ ( w ) ∼ ˙ e − ζ ( z ) s = 0 ∑ ∞ z ( N + 1 ) s + N + p ( − 1 ) s + 1 E ˉ s , p − Θ ( z ) S ˉ N + 1 N + p (126)
where we defined
E ˉ s , p = ( N + 1 ) s Γ ( N + p N + 1 + s ) ( 2 i ) s + 1 Γ ( N + p N + 1 ) , S ˉ N + p N + 1 = π e − i π 2 N + p N + 1 Γ ( N + p N + 1 ) ( N + 1 2 ) N + p N + 1 (127) \bar{E}_{s,p}=\frac{(N+1)^s\Gamma\left(\frac{N+p}{N+1}+s\right)}{(2i)^{s+1}\Gamma\left(\frac{N+p}{N+1}\right)},\qquad
\bar{S}_{\frac{N+p}{N+1}}=\frac{\pi e^{-\frac{i\pi}{2}\frac{N+p}{N+1}}}{\Gamma\left(\frac{N+p}{N+1}\right)\left(\frac{N+1}{2}\right)^{\frac{N+p}{N+1}}} \tag*{(127)} E ˉ s , p = ( 2 i ) s + 1 Γ ( N + 1 N + p ) ( N + 1 ) s Γ ( N + 1 N + p + s ) , S ˉ N + 1 N + p = Γ ( N + 1 N + p ) ( 2 N + 1 ) N + 1 N + p π e − 2 iπ N + 1 N + p (127)
and Θ ( z ) \Theta(z) Θ ( z ) in eq. (110) .
With the above, the simplification of the right hand sides of eqs. (120) and (121) may be understood in terms of the following linear maps on P ( z ) \mathcal{P}(z) P ( z ) :
B ± [ f ( z ) ] ≡ B ± ( z ) f ( z ) , (128) \mathcal{B}_{\pm}[f(z)]\equiv\mathcal{B}_{\pm}(z)f(z), \tag*{(128)} B ± [ f ( z )] ≡ B ± ( z ) f ( z ) , (128)
where P ≥ 2 ( z ) \mathcal{P}_{\geq2}(z) P ≥ 2 ( z ) is the subspace of P ( z ) \mathcal{P}(z) P ( z ) with vanishing terms of order z − 1 z^{-1} z − 1 .
∫ ∞ z d w e ζ ( w ) f ( w ) ≡ e ζ ( z ) E + [ f ( z ) ] (129) \int_{\infty}^{z}dw\,e^{\zeta(w)}f(w)\equiv e^{\zeta(z)}\mathcal{E}_{+}[f(z)] \tag*{(129)} ∫ ∞ z d w e ζ ( w ) f ( w ) ≡ e ζ ( z ) E + [ f ( z )] (129)
∫ ∞ z d w e − ζ ( w ) f ( w ) ≡ e − ζ ( z ) E − [ f ( z ) ] + Θ ( z ) S [ f ] (130) \int_{\infty}^{z}dw\,e^{-\zeta(w)}f(w)\equiv e^{-\zeta(z)}\mathcal{E}_{-}[f(z)]+\Theta(z)\mathcal{S}[f] \tag*{(130)} ∫ ∞ z d w e − ζ ( w ) f ( w ) ≡ e − ζ ( z ) E − [ f ( z )] + Θ ( z ) S [ f ] (130)
where the integrals on f ( z ) ∈ P ( z ) f(z)\in\mathcal{P}(z) f ( z ) ∈ P ( z ) are to be understood as evaluated term-wise using eq. (125) and (126) .21
∫ ∞ z d w f ( w ) ≡ I [ f ( z ) ] . (131) \int_{\infty}^{z}dw\,f(w)\equiv\mathcal{I}[f(z)]. \tag*{(131)} ∫ ∞ z d w f ( w ) ≡ I [ f ( z )] . (131)
Explicit expressions for the above are provided in appendix G.
In terms of the above, the integrals of eq. (120) and (121) can be expressed as
∫ ∞ z d w e ζ ( w ) B − ( w ) β ˉ ( w ) ∼ ˙ T 1 Θ ( z ) e ζ ( z ) E + [ B − ( z ) ] + T 1 Θ ( z ) e ζ ( z ) E + B − [ λ 2 ( z ) ] + I B − [ μ ˉ 2 ( z ) ] (132) \int_{\infty}^{z}dw\,e^{\zeta(w)}\mathcal{B}_{-}(w)\bar{\beta}(w)\mathrel{\dot{\sim}}T_{1}\Theta(z)e^{\zeta(z)}\mathcal{E}_{+}[\mathcal{B}_{-}(z)]+T_{1}\Theta(z)e^{\zeta(z)}\mathcal{E}_{+}\mathcal{B}_{-}[\lambda_{2}(z)]+\mathcal{I}\mathcal{B}_{-}[\bar{\mu}_{2}(z)] \tag*{(132)} ∫ ∞ z d w e ζ ( w ) B − ( w ) β ˉ ( w ) ∼ ˙ T 1 Θ ( z ) e ζ ( z ) E + [ B − ( z )] + T 1 Θ ( z ) e ζ ( z ) E + B − [ λ 2 ( z )] + I B − [ μ ˉ 2 ( z )] (132)
∫ ∞ z d w e − ζ ( w ) B + ( w ) α ( w ) ∼ ˙ e − ζ ( z ) E − [ B + ( z ) + B + [ λ 1 ( z ) ] ] + Θ ( z ) S [ B + ( z ) + B + [ λ 1 ( z ) ] ] + T 1 Θ ( z ) I [ B + [ μ ˉ 1 ( z ) ] ] . (133) \int_{\infty}^{z}dw\,e^{-\zeta(w)}\mathcal{B}_{+}(w)\alpha(w)\mathrel{\dot{\sim}}e^{-\zeta(z)}\mathcal{E}_{-}[\mathcal{B}_{+}(z)+\mathcal{B}_{+}[\lambda_{1}(z)]]+\Theta(z)\mathcal{S}[\mathcal{B}_{+}(z)+\mathcal{B}_{+}[\lambda_{1}(z)]]+T_{1}\Theta(z)\mathcal{I}[\mathcal{B}_{+}[\bar{\mu}_{1}(z)]]. \tag*{(133)} ∫ ∞ z d w e − ζ ( w ) B + ( w ) α ( w ) ∼ ˙ e − ζ ( z ) E − [ B + ( z ) + B + [ λ 1 ( z )]] + Θ ( z ) S [ B + ( z ) + B + [ λ 1 ( z )]] + T 1 Θ ( z ) I [ B + [ μ ˉ 1 ( z )]] . (133)
Comparing with the left hand sides which are of the form eq. (122) and eq. (123) we find
λ 1 ( z ) = I B − [ μ ˉ 2 ( z ) ] (134) \lambda_{1}(z)=\mathcal{I}\mathcal{B}_{-}[\bar{\mu}_{2}(z)] \tag*{(134)} λ 1 ( z ) = I B − [ μ ˉ 2 ( z )] (134)
λ 2 ( z ) = I B + [ μ ˉ 1 ( z ) ] (135) \lambda_{2}(z)=\mathcal{I}\mathcal{B}_{+}[\bar{\mu}_{1}(z)] \tag*{(135)} λ 2 ( z ) = I B + [ μ ˉ 1 ( z )] (135)
μ ˉ 1 ( z ) = E + [ B − ( z ) + B − [ λ 2 ( z ) ] ] (136) \bar{\mu}_{1}(z)=\mathcal{E}_{+}[\mathcal{B}_{-}(z)+\mathcal{B}_{-}[\lambda_{2}(z)]] \tag*{(136)} μ ˉ 1 ( z ) = E + [ B − ( z ) + B − [ λ 2 ( z )]] (136)
μ ˉ 2 ( z ) = E − [ D ( z ) ] (137) \bar{\mu}_{2}(z)=\mathcal{E}_{-}[D(z)] \tag*{(137)} μ ˉ 2 ( z ) = E − [ D ( z )] (137)
T 1 = S [ D ( z ) ] (138) T_{1}=\mathcal{S}[D(z)] \tag*{(138)} T 1 = S [ D ( z )] (138)
where
D ( z ) ≡ B + ( z ) + B + [ λ 1 ( z ) ] . (139) D(z)\equiv\mathcal{B}_{+}(z)+\mathcal{B}_{+}[\lambda_{1}(z)]. \tag*{(139)} D ( z ) ≡ B + ( z ) + B + [ λ 1 ( z )] . (139)
We note that the above relations recursively solve for λ m ( 1 , 2 ) \lambda_{m}^{(1,2)} λ m ( 1 , 2 ) and μ m ( 1 , 2 ) \mu_{m}^{(1,2)} μ m ( 1 , 2 ) in terms of B m ( ± ) \mathcal{B}_{m}^{(\pm)} B m ( ± ) , and allow T 1 T_{1} T 1 to be computed as a known function of these. In particular, note composing (134) and (137) we obtain
D ( z ) = B + ( z ) + B + I B − E − [ D ( z ) ] (140) D(z)=\mathcal{B}_{+}(z)+\mathcal{B}_{+}\mathcal{I}\mathcal{B}_{-}\mathcal{E}_{-}[D(z)] \tag*{(140)} D ( z ) = B + ( z ) + B + I B − E − [ D ( z )] (140)
which gives T 1 T_1 T 1 through eq. (138) and in turn the Stokes constant through eq. (119) . We will discuss the detailed structure of the power series representation for T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) in the next subsection.
Explicit recursion relation and its solution
The goal of this subsection is to derive a closed form expression for T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) to leading order of k 2 k^2 k 2 . Specifically, we will derive eqs. (152) and (153) , the T 1 ( 0 ) T_1^{(0)} T 1 ( 0 ) and T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) coefficients in the soft momentum expansion of T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) about the topological limit k = 0 k=0 k = 0 :
T 1 ( k 2 ) ≡ T 1 ( 0 ) + T 1 ( 2 ) k 2 + O ( k 4 ) . (141) T_1(k^2) \equiv T_1^{(0)} + T_1^{(2)} k^2 + O(k^4). \tag*{(141)} T 1 ( k 2 ) ≡ T 1 ( 0 ) + T 1 ( 2 ) k 2 + O ( k 4 ) . (141)
We have seen in eq. (138) that T 1 T_1 T 1 is given by the Stokes functional S \mathcal{S} S acting on the D ( z ) D(z) D ( z ) function with D ( z ) D(z) D ( z ) determined by the recursion relation (140) . In this subsection, we will solve a rescaled D ( z ) D(z) D ( z ) recursion relation (see eq. (149) ) in the { z − m } \{z^{-m}\} { z − m } basis of the functional space P ( z ) \mathcal{P}(z) P ( z ) . Based on this solution, we will derive closed form expressions for T 1 ( 0 ) T_1^{(0)} T 1 ( 0 ) and T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) as functions of the dispersion index N = n / 2 N=n/2 N = n /2 . They will be used in section 5.1 for deriving soft corrections for the Bogoliubov coefficient β k \beta_k β k .
To start, we expand the functions D ( z ) D(z) D ( z ) and B ± ( z ) B_{\pm}(z) B ± ( z ) in the { z − m } \{z^{-m}\} { z − m } basis of the functional space P ( z ) \mathcal{P}(z) P ( z ) as
D ( z ) ≡ ∑ m = 1 ∞ D m z − m , B + ( z ) ≡ ∑ m = 1 ∞ B m ( + ) z − m , B − ( z ) ≡ ∑ m = 1 ∞ B m ( − ) z − m . (142) D(z) \equiv\sum_{m=1}^{\infty}D_m z^{-m}, \qquad B_{+}(z) \equiv\sum_{m=1}^{\infty}B_m^{(+)}z^{-m}, \qquad B_{-}(z) \equiv\sum_{m=1}^{\infty}B_m^{(-)}z^{-m}. \tag*{(142)} D ( z ) ≡ m = 1 ∑ ∞ D m z − m , B + ( z ) ≡ m = 1 ∑ ∞ B m ( + ) z − m , B − ( z ) ≡ m = 1 ∑ ∞ B m ( − ) z − m . (142)
Next, we reduce the linear integral equation (140) for D ( z ) D(z) D ( z ) by working out the explicit representation of the object B + B − E − [ D ( z ) ] \mathcal{B}_{+}\mathcal{B}_{-}\mathcal{E}_{-}[D(z)] B + B − E − [ D ( z )] . Using the explicit representations of the linear operators given by eqs. (G.4)–(G.8), we obtain
E − [ D ( z ) ] = ∑ m = 1 ∞ 1 z m + N ∑ p = 0 m − 1 D p + 1 ( − 1 ) m − p − 1 N + 1 + 1 E ˉ m − p − 1 N + 1 , p + 1 δ ( m − p − 1 N + 1 ) = ∑ l = N + 1 ∞ 1 z l ∑ s = 0 l − N − 1 F l − N , s D s + 1 . (143) \begin{aligned}
\mathcal{E}_{-}[D(z)]={}&\sum_{m=1}^{\infty}\frac{1}{z^{m+N}}\sum_{p=0}^{m-1}D_{p+1}(-1)^{\frac{m-p-1}{N+1}+1}\bar{E}_{\frac{m-p-1}{N+1},p+1}\delta\left(\frac{m-p-1}{N+1}\right)\\
={}&\sum_{l=N+1}^{\infty}\frac{1}{z^l}\sum_{s=0}^{l-N-1}F_{l-N,s}D_{s+1}. \tag*{(143)}
\end{aligned} E − [ D ( z )] = = m = 1 ∑ ∞ z m + N 1 p = 0 ∑ m − 1 D p + 1 ( − 1 ) N + 1 m − p − 1 + 1 E ˉ N + 1 m − p − 1 , p + 1 δ ( N + 1 m − p − 1 ) l = N + 1 ∑ ∞ z l 1 s = 0 ∑ l − N − 1 F l − N , s D s + 1 . (143)
where
F m , s ≡ ( − 1 ) m − s − 1 N + 1 + 1 E ˉ m − s − 1 N + 1 , s + 1 δ ( m − s − 1 N + 1 ) . (144) F_{m,s}\equiv(-1)^{\frac{m-s-1}{N+1}+1}\bar{E}_{\frac{m-s-1}{N+1},s+1}\delta\left(\frac{m-s-1}{N+1}\right). \tag*{(144)} F m , s ≡ ( − 1 ) N + 1 m − s − 1 + 1 E ˉ N + 1 m − s − 1 , s + 1 δ ( N + 1 m − s − 1 ) . (144)
It follows that
B + B − E − [ D ( z ) ] = − ∑ p = 2 ∞ 1 z p ∑ m + l = p B m ( + ) 1 l ( ∑ q + r = l + 1 B q ( − ) ∑ s = 0 r − N − 1 F r − N , s D s + 1 ) . (145) \mathcal{B}_{+}\mathcal{B}_{-}\mathcal{E}_{-}[D(z)]=-\sum_{p=2}^{\infty}\frac{1}{z^p}\sum_{m+l=p}B_m^{(+)}\frac{1}{l}\left(\sum_{q+r=l+1}B_q^{(-)}\sum_{s=0}^{r-N-1}F_{r-N,s}D_{s+1}\right). \tag*{(145)} B + B − E − [ D ( z )] = − p = 2 ∑ ∞ z p 1 m + l = p ∑ B m ( + ) l 1 q + r = l + 1 ∑ B q ( − ) s = 0 ∑ r − N − 1 F r − N , s D s + 1 . (145)
With the above results, comparing the coefficients of the { z − m } \{z^{-m}\} { z − m } basis in eq. (140) , we get the explicit recursion relation for D ( z ) D(z) D ( z ) :
D p + 1 = { B p + 1 ( + ) 0 ≤ p ≤ N , B p + 1 ( + ) − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 B p − m + 1 ( + ) B m − l + 1 ( − ) m F l − N , s D s + 1 p ≥ N + 1. (146) D_{p+1}=
\begin{cases}
B_{p+1}^{(+)} & 0\leq p\leq N,\\
B_{p+1}^{(+)}-\sum_{m=N+1}^{p}\sum_{l=N+1}^{m}\sum_{s=0}^{l-N-1}\dfrac{B_{p-m+1}^{(+)}B_{m-l+1}^{(-)}}{m}F_{l-N,s}D_{s+1} & p\geq N+1.
\end{cases} \tag*{(146)} D p + 1 = ⎩ ⎨ ⎧ B p + 1 ( + ) B p + 1 ( + ) − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 m B p − m + 1 ( + ) B m − l + 1 ( − ) F l − N , s D s + 1 0 ≤ p ≤ N , p ≥ N + 1. (146)
where E ˉ s , p \bar{E}_{s,p} E ˉ s , p and δ \delta δ are defined by eq. (F.12) and eq. (G.3), respectively.
Now, it is convenient to define a rescaled variable for D ( z ) D(z) D ( z ) encoded by D p D_p D p . Use eqs. (138) and (G.6) to find a series expression for T 1 T_1 T 1 :
T 1 = − ∑ m = 1 ∞ D m S ˉ N + m N + 1 = − ∑ m = 1 ∞ G m (147) T_1=-\sum_{m=1}^{\infty}D_m\bar{S}_{\frac{N+m}{N+1}}=-\sum_{m=1}^{\infty}G_m \tag*{(147)} T 1 = − m = 1 ∑ ∞ D m S ˉ N + 1 N + m = − m = 1 ∑ ∞ G m (147)
with S ˉ \bar{S} S ˉ defined by eq. (F.12) coming from S S S operator acting on z − m z^{-m} z − m and
G m ≡ D m S ˉ N + m N + 1 , G ( z ) ≡ ∑ m = 1 ∞ G m z − m (148) G_m\equiv D_m\bar{S}_{\frac{N+m}{N+1}},\qquad G(z)\equiv\sum_{m=1}^{\infty}G_mz^{-m} \tag*{(148)} G m ≡ D m S ˉ N + 1 N + m , G ( z ) ≡ m = 1 ∑ ∞ G m z − m (148)
with the G ( z ) G(z) G ( z ) and G m G_m G m definitions identified as naturally rescaled D ( z ) D(z) D ( z ) and D p D_p D p . The explicit recursion relation for G ( z ) G(z) G ( z ) is then obtained by rescaling each D p D_p D p term in eq. (146) by S ˉ ( N + p ) / ( N + 1 ) \bar{S}_{(N+p)/(N+1)} S ˉ ( N + p ) / ( N + 1 ) according to eq. (148) , after which
G p + 1 = { B p + 1 ( + ) S ˉ N + p + 1 N + 1 0 ≤ p ≤ N , B p + 1 ( + ) S ˉ N + p + 1 N + 1 − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s G s + 1 p ≥ N + 1. (149) G_{p+1}=
\begin{cases}
B_{p+1}^{(+)}\bar{S}_{\frac{N+p+1}{N+1}} & 0\leq p\leq N,\\
B_{p+1}^{(+)}\bar{S}_{\frac{N+p+1}{N+1}}-\sum_{m=N+1}^{p}\sum_{l=N+1}^{m}\sum_{s=0}^{l-N-1}K_{p;m,l,s}G_{s+1} & p\geq N+1.
\end{cases} \tag*{(149)} G p + 1 = { B p + 1 ( + ) S ˉ N + 1 N + p + 1 B p + 1 ( + ) S ˉ N + 1 N + p + 1 − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s G s + 1 0 ≤ p ≤ N , p ≥ N + 1. (149)
K p ; m , l , s = B p − m + 1 ( + ) B m − l + 1 ( − ) m l S ˉ N + p + 1 N + 1 S ˉ N + l + 1 N + 1 δ ( l − s N + 1 − 1 ) (150) K_{p;m,l,s}=\frac{B_{p-m+1}^{(+)}B_{m-l+1}^{(-)}}{ml}\frac{\bar{S}_{\frac{N+p+1}{N+1}}}{\bar{S}_{\frac{N+l+1}{N+1}}}\delta\left(\frac{l-s}{N+1}-1\right) \tag*{(150)} K p ; m , l , s = m l B p − m + 1 ( + ) B m − l + 1 ( − ) S ˉ N + 1 N + l + 1 S ˉ N + 1 N + p + 1 δ ( N + 1 l − s − 1 ) (150)
where the definitions for S ˉ y \bar{S}_y S ˉ y and δ ( X ) \delta(X) δ ( X ) can be found in (F.12) and (G.3).
In the soft momentum limit k → 0 k\to0 k → 0 , each G p G_p G p term splits into the topological limit term G p ( 0 ) G_p^{(0)} G p ( 0 ) and the soft correction term G p ( 2 ) G_p^{(2)} G p ( 2 ) :
G p = G p ( 0 ) + G p ( 2 ) k 2 + O ( k 4 ) . (151) G_p=G_p^{(0)}+G_p^{(2)}k^2+O(k^4). \tag*{(151)} G p = G p ( 0 ) + G p ( 2 ) k 2 + O ( k 4 ) . (151)
Then, we show in appendix § 17 .§ 1 that the explicit recursion relation (149) also splits into two sets of recursion relations for G p ( 0 ) G_p^{(0)} G p ( 0 ) and G p ( 2 ) G_p^{(2)} G p ( 2 ) , respectively. We solve the split recursion relations in appendix § 17 .§ 3 to obtain closed form expressions for the partial sums of G p ( 0 ) G_p^{(0)} G p ( 0 ) and G p ( 2 ) G_p^{(2)} G p ( 2 ) (see eqs. (J.26) and (J.76)). This in turn yields closed form expressions for the T 1 ( 0 ) T_1^{(0)} T 1 ( 0 ) and T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) coefficients (see eqs. (J.31) and (J.102)):
T 1 ( 0 ) = − ∑ p = 1 ∞ G p ( 0 ) = 2 i cos [ π 2 ( N + 1 ) ] (152) T_1^{(0)}=-\sum_{p=1}^{\infty}G_p^{(0)}=2i\cos\left[\frac{\pi}{2(N+1)}\right] \tag*{(152)} T 1 ( 0 ) = − p = 1 ∑ ∞ G p ( 0 ) = 2 i cos [ 2 ( N + 1 ) π ] (152)
T 1 ( 2 ) = − ∑ p = 1 ∞ G p ( 2 ) = − i [ 1 i ( N + 1 ) ] 2 N N + 1 π 3 / 2 Γ [ N − 1 2 ( N + 1 ) ] Γ ( N N + 1 ) Γ [ 2 N + 1 2 ( N + 1 ) ] Γ [ 2 N − 1 2 ( N + 1 ) ] (153) T_1^{(2)}=-\sum_{p=1}^{\infty}G_p^{(2)}=-i\left[\frac{1}{i(N+1)}\right]^{\frac{2N}{N+1}}\pi^{3/2}\frac{\Gamma\left[\frac{N-1}{2(N+1)}\right]}{\Gamma\left(\frac{N}{N+1}\right)\Gamma\left[\frac{2N+1}{2(N+1)}\right]\Gamma\left[\frac{2N-1}{2(N+1)}\right]} \tag*{(153)} T 1 ( 2 ) = − p = 1 ∑ ∞ G p ( 2 ) = − i [ i ( N + 1 ) 1 ] N + 1 2 N π 3/2 Γ ( N + 1 N ) Γ [ 2 ( N + 1 ) 2 N + 1 ] Γ [ 2 ( N + 1 ) 2 N − 1 ] Γ [ 2 ( N + 1 ) N − 1 ] (153)
where we have used eqs. (141) and (147) .
These two closed form expressions lead to a closed form expression for the first Stokes constant in the soft limit
S 1 ( k ) = e − 2 i θ ˉ ( k ) T 1 ( k 2 ) = e − 2 i θ ˉ ( k ) [ T 1 ( 0 ) + T 1 ( 2 ) k 2 + O ( k 4 ) ] . (154) S_1(k)=e^{-2i\bar{\theta}(k)}T_1(k^2)=e^{-2i\bar{\theta}(k)}\left[T_1^{(0)}+T_1^{(2)}k^2+O(k^4)\right]. \tag*{(154)} S 1 ( k ) = e − 2 i θ ˉ ( k ) T 1 ( k 2 ) = e − 2 i θ ˉ ( k ) [ T 1 ( 0 ) + T 1 ( 2 ) k 2 + O ( k 4 ) ] . (154)
We have seen in eq. (48) that the particle production coefficient β k \beta_k β k is equal to the 21 entry of a matrix product expressed in terms of Stokes constants, and in eqs. (70) to (3.24) that all the Stokes constants are related to the first Stokes constant via a discrete rotational symmetry. These suggest that a closed form expression for β k \beta_k β k in the soft limit is obtainable. Deriving the closed form expression for β k \beta_k β k to O ( k 2 ) O(k^2) O ( k 2 ) order is therefore the task of section 5.1.
Soft corrections to Bogoliubov $\beta_k`
Based on the framework established in sections 3 and 4, we are now ready to compute soft corrections to the Bogoliubov β k \beta_k β k using eq. (48) . In section 5.1, we will show that the leading order expansion of β k \beta_k β k about the topological limit k = 0 k=0 k = 0 is quadratic in k k k , i.e.,
β k = β k = 0 + c 1 k 2 + O ( k 4 ) , (155) \beta_k = \beta_{k=0} + c_1 k^2 + O(k^4), \tag*{(155)} β k = β k = 0 + c 1 k 2 + O ( k 4 ) , (155)
with β k = 0 = − i cot ( π / ( n + 2 ) ) \beta_{k=0}=-i\cot(\pi/(n+2)) β k = 0 = − i cot ( π / ( n + 2 )) the topological limit result [9 ] and c 1 c_1 c 1 a purely imaginary constant as required by the time-reversal symmetry (cf. eq. (85) ). We first evaluate c 1 c_1 c 1 analytically via eq. (48) , utilizing S 1 ( k ) = exp [ − 2 i θ ˉ ( k ) ] T 1 ( k 2 ) S_1(k)=\exp[-2i\bar{\theta}(k)]T_1(k^2) S 1 ( k ) = exp [ − 2 i θ ˉ ( k )] T 1 ( k 2 ) where T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) to leading order O ( k 2 ) O(k^2) O ( k 2 ) is given by eqs. (141) , (4.67) and (4.68). We then compare this to a direct computation of β k \beta_k β k using the Wronskian projection (16) , obtained by numerically solving the mode equation (29) and find good agreement.
Surprisingly, for n ≥ 4 n\geq4 n ≥ 4 , the non-analytic k k k factors Δ j ( k ) \Delta_j(k) Δ j ( k ) arising from branch cuts and the non-analytic residue θ ˉ ( k ) \bar{\theta}(k) θ ˉ ( k ) in the Stokes constants explicitly cancel out. For n = 2 n=2 n = 2 , we will see from eq. (48) that β k \beta_k β k only receives contributions from Δ 1 ( k ) \Delta_1(k) Δ 1 ( k ) , which is analytic in k k k .
Analytic $\beta_k`
As we have shown in section 2.2, propagating the WKB amplitudes on the lower half plane and utilizing Stokes phenomena leads to the expression (48) where β k \beta_k β k is written as the 21 entry of a product of propagators across Stokes sectors and branch cut matrices. In this subsection, we will further simplify the above expression, after which β k \beta_k β k will be given as a product of matrices whose entries are the power series T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) with the arguments rotated. We will also use the notations introduced in appendix C.
We first consider the cases where n ≥ 4 n\geq4 n ≥ 4 , i.e., N ≥ 2 N\geq2 N ≥ 2 . Let j j j be an integer in the range 1 ≤ j ≤ N 1\leq j\leq N 1 ≤ j ≤ N . As a first step, note that the branch cut matrices B N + j ( k ) \mathcal{B}_{N+j}(k) B N + j ( k ) defined in eq. (45) can be rewritten as a product of rotational representation matrices B B B and R N + j ( k ) \mathcal{R}_{N+j}(k) R N + j ( k ) defined in eq. (68) as
B N + j ( k ) = − i R [ Δ N + j ( k ) ] B . (156) \mathcal{B}_{N+j}(k)=-i\mathcal{R}[\Delta_{N+j}(k)]B. \tag*{(156)} B N + j ( k ) = − i R [ Δ N + j ( k )] B . (156)
Next, we rotate the propagators U N + 2 + j ( k ) U_{N+2+j}(k) U N + 2 + j ( k ) using eqs. (72) and (C.2)
U N + 2 + j ( k ) = U + [ S N + 2 + j ( k ) ] = U + [ e 4 i Φ ~ j S ~ 1 ( N + 2 + j ) ( k ) ] = R [ Φ ~ j ] U + [ S ~ 1 ( N + 2 + j ) ( k ) ] R − 1 [ Φ ~ j ] , (157) U_{N+2+j}(k)=U_+[S_{N+2+j}(k)]=U_+[e^{4i\widetilde{\Phi}_j}\widetilde{S}_1^{(N+2+j)}(k)]=\mathcal{R}[\widetilde{\Phi}_j]U_+[\widetilde{S}_1^{(N+2+j)}(k)]\mathcal{R}^{-1}[\widetilde{\Phi}_j], \tag*{(157)} U N + 2 + j ( k ) = U + [ S N + 2 + j ( k )] = U + [ e 4 i Φ j S 1 ( N + 2 + j ) ( k )] = R [ Φ j ] U + [ S 1 ( N + 2 + j ) ( k )] R − 1 [ Φ j ] , (157)
where S ~ 1 ( m ) ( k ) ≡ S 1 ( γ − N ( m − 1 ) k ) \widetilde{S}_1^{(m)}(k) \equiv S_1(\gamma^{-N(m-1)}k) S 1 ( m ) ( k ) ≡ S 1 ( γ − N ( m − 1 ) k ) with 1 ≤ m ≤ 2 N + 2 1 \le m \le2N+2 1 ≤ m ≤ 2 N + 2 denotes the first Stokes constant with the argument k k k rotated. From the decomposition (119) and the form of the non-analytic residue (100) , we see that
S ~ 1 ( m ) ( k ) = e − 2 i θ ˉ ( γ − N ( m − 1 ) k ) T 1 ( γ − 2 N ( m − 1 ) k 2 ) = e − 2 i ( − 1 ) m − 1 θ ˉ ( k ) T 1 ( γ − 2 N ( m − 1 ) k 2 ) . (158) \widetilde{S}_1^{(m)}(k)=e^{-2i\bar{\theta}(\gamma^{-N(m-1)}k)}T_1(\gamma^{-2N(m-1)}k^2)=e^{-2i(-1)^{m-1}\bar{\theta}(k)}T_1(\gamma^{-2N(m-1)}k^2). \tag*{(158)} S 1 ( m ) ( k ) = e − 2 i θ ˉ ( γ − N ( m − 1 ) k ) T 1 ( γ − 2 N ( m − 1 ) k 2 ) = e − 2 i ( − 1 ) m − 1 θ ˉ ( k ) T 1 ( γ − 2 N ( m − 1 ) k 2 ) . (158)
Combining eqs. (156) and (157) , we find
U N + 2 + j B N + j = − i R [ Δ N + j ] U + [ e − 4 i ( Δ N + j − Φ ~ j ) S ~ 1 ( N + 2 + j ) ] B , (159) U_{N+2+j}\mathcal{B}_{N+j}=-i\mathcal{R}[\Delta_{N+j}]U_+\left[e^{-4i(\Delta_{N+j}-\widetilde{\Phi}_j)}\widetilde{S}_1^{(N+2+j)}\right]B, \tag*{(159)} U N + 2 + j B N + j = − i R [ Δ N + j ] U + [ e − 4 i ( Δ N + j − Φ j ) S 1 ( N + 2 + j ) ] B , (159)
where we have used eq. (C.2). Subsequently, we may simplify the propagator U ( + ∞ , − ∞ ) U(+\infty,-\infty) U ( + ∞ , − ∞ ) using the properties B R [ x ] B = R [ − x ] B\mathcal{R}[x]B=\mathcal{R}[-x] B R [ x ] B = R [ − x ] and R [ x ] U ± [ S ] R − 1 [ x ] = U ± [ e ± 4 i x S ] \mathcal{R}[x]U_\pm[S]\mathcal{R}^{-1}[x]=U_\pm[e^{\pm4ix}S] R [ x ] U ± [ S ] R − 1 [ x ] = U ± [ e ± 4 i x S ] to find
U 0 ( k , η + ∞ , η − ∞ ) = [ ∏ j = 1 N ( U N + 2 + j B N + j ) ] U N + 2 = ( − 1 ) N i N R [ − Δ ~ N ] ∏ j = 1 N ( U + [ e 4 i ( − 1 ) j Δ ~ N S ~ 1 ( N + 2 + j ) ] B ) U + [ e 4 i Δ ~ N S ~ 1 ( N + 2 ) ] , (160) \begin{aligned}
U_0(k,\eta_{+\infty},\eta_{-\infty})=\left[\prod_{j=1}^{N}(U_{N+2+j}\mathcal{B}_{N+j})\right]U_{N+2} \\
&=(-1)^N i^N\mathcal{R}[-\widetilde{\Delta}_N]\prod_{j=1}^{N}\left(U_+\left[e^{4i(-1)^j\widetilde{\Delta}_N}\widetilde{S}_1^{(N+2+j)}\right]B\right)U_+\left[e^{4i\widetilde{\Delta}_N}\widetilde{S}_1^{(N+2)}\right], \tag*{(160)}
\end{aligned} U 0 ( k , η + ∞ , η − ∞ ) = [ j = 1 ∏ N ( U N + 2 + j B N + j ) ] U N + 2 = ( − 1 ) N i N R [ − Δ N ] j = 1 ∏ N ( U + [ e 4 i ( − 1 ) j Δ N S 1 ( N + 2 + j ) ] B ) U + [ e 4 i Δ N S 1 ( N + 2 ) ] , (160)
where in the last line we have used − Δ ~ j − Φ ~ j = − ( − 1 ) j Δ ~ N -\widetilde{\Delta}_j-\widetilde{\Phi}_j=-(-1)^j\widetilde{\Delta}_N − Δ j − Φ j = − ( − 1 ) j Δ N and where ∏ j = 1 N \prod_{j=1}^{N} ∏ j = 1 N denotes the matrix product from right to left with j j j ascending. The Bogoliubov β k \beta_k β k is then given by the 21 entry of the above product as
β k = ( − 1 ) N i N e 2 i Δ ~ N [ ∏ j = 1 N ( U + [ e 4 i ( − 1 ) j Δ ~ N S ~ 1 ( N + 2 + j ) ] B ) ] 21 . (161) \beta_k=(-1)^N i^N e^{2i\widetilde{\Delta}_N}\left[\prod_{j=1}^{N}\left(U_+\left[e^{4i(-1)^j\widetilde{\Delta}_N}\widetilde{S}_1^{(N+2+j)}\right]B\right)\right]_{21}. \tag*{(161)} β k = ( − 1 ) N i N e 2 i Δ N [ j = 1 ∏ N ( U + [ e 4 i ( − 1 ) j Δ N S 1 ( N + 2 + j ) ] B ) ] 21 . (161)
In appendix § 16 , we detail the k k k non-analyticity cancellation and simplify the above result to
β k = ( − 1 ) N i N [ ∏ j = 1 N ( U + [ T ~ 1 ( N + 2 + j ) ( k 2 ) ] B ) ] 21 , (162) \beta_k=(-1)^N i^N\left[\prod_{j=1}^{N}\left(U_+\left[\widetilde{T}_1^{(N+2+j)}(k^2)\right]B\right)\right]_{21}, \tag*{(162)} β k = ( − 1 ) N i N [ j = 1 ∏ N ( U + [ T 1 ( N + 2 + j ) ( k 2 ) ] B ) ] 21 , (162)
where
T ~ 1 ( N + 2 + j ) ( k 2 ) ≡ T 1 ( γ − 2 N ( N + 1 + j ) k 2 ) (163) \widetilde{T}_1^{(N+2+j)}(k^2)\equiv T_1(\gamma^{-2N(N+1+j)}k^2) \tag*{(163)} T 1 ( N + 2 + j ) ( k 2 ) ≡ T 1 ( γ − 2 N ( N + 1 + j ) k 2 ) (163)
are rotated power series. Now the Bogoliubov β k \beta_k β k is written as a finite product of T ~ 1 ( j ) ( k 2 ) \widetilde{T}_1^{(j)}(k^2) T 1 ( j ) ( k 2 ) . Hence, the soft corrections to the Bogoliubov β k \beta_k β k for n ≥ 4 n \ge4 n ≥ 4 (i.e., N ≥ 2 N \ge2 N ≥ 2 ) are analytic in k k k and specifically, the leading order correction is quadratic.
For n = 2 n=2 n = 2 (i.e., N = 1 N=1 N = 1 ), the derivations up to eq. (161) are still valid, giving
β k = − i e 2 i Δ ~ 1 [ U + [ e − 4 i Δ ~ 1 S ~ 1 ( 4 ) ] B ] 21 = − i e 2 i Δ 1 , (164) \beta_k=-ie^{2i\widetilde{\Delta}_1}\left[U_+\left[e^{-4i\widetilde{\Delta}_1}\widetilde{S}_1^{(4)}\right]B\right]_{21}=-ie^{2i\Delta_1}, \tag*{(164)} β k = − i e 2 i Δ 1 [ U + [ e − 4 i Δ 1 S 1 ( 4 ) ] B ] 21 = − i e 2 i Δ 1 , (164)
which is independent of the first Stokes constant. For N = 1 N=1 N = 1 , we have Δ 1 = i π k 2 / 4 \Delta_1=i\pi k^2/4 Δ 1 = iπ k 2 /4 as can be seen from eq. (B.2), then
β k = − i e − π k 2 2 , (165) \beta_k=-ie^{-\frac{\pi k^2}{2}}, \tag*{(165)} β k = − i e − 2 π k 2 , (165)
which is also analytic in k k k , and agrees with the solvable model in the literature (see e.g. [21 ] ).
With the above, the soft corrections to Bogoliubov β k \beta_k β k to quadratic order for arbitrary n ∈ e v e n n \in\mathrm{even} n ∈ even are given by
β k ( n ) = − i cot ( π n + 2 ) + c 1 ( n ) k 2 + O ( k 4 ) (166) \beta_k(n)=-i\cot\left(\frac{\pi}{n+2}\right)+c_1(n)k^2+O(k^4) \tag*{(166)} β k ( n ) = − i cot ( n + 2 π ) + c 1 ( n ) k 2 + O ( k 4 ) (166)
with c 1 ( n ) c_1(n) c 1 ( n ) derived in appendix J using eq. (162) as
c 1 ( n ) = i 2 4 n + 2 − 1 π ( n + 2 2 ) 2 − 4 n + 2 cos ( π n + 2 ) Γ ( 1 n + 2 ) [ Γ ( 2 n + 2 ) ] 2 Γ ( 3 n + 2 ) Γ ( 4 n + 2 ) . (167) c_1(n)=i\frac{2^{\frac{4}{n+2}-1}}{\pi\left(\frac{n+2}{2}\right)^{2-\frac{4}{n+2}}}\cos\left(\frac{\pi}{n+2}\right)\frac{\Gamma\left(\frac{1}{n+2}\right)\left[\Gamma\left(\frac{2}{n+2}\right)\right]^2\Gamma\left(\frac{3}{n+2}\right)}{\Gamma\left(\frac{4}{n+2}\right)}. \tag*{(167)} c 1 ( n ) = i π ( 2 n + 2 ) 2 − n + 2 4 2 n + 2 4 − 1 cos ( n + 2 π ) Γ ( n + 2 4 ) Γ ( n + 2 1 ) [ Γ ( n + 2 2 ) ] 2 Γ ( n + 2 3 ) . (167)
This is the main mathematical result of this paper. To verify eq. (167) , we compute β k \beta_k β k numerically for n = 4 , 6 , 8 n=4,6,8 n = 4 , 6 , 8 , and 10 10 10 by solving the mode equation (29) with the WKB boundary conditions (13) and using the Wronskian projection (16) . The object ( β k n u m e r i c a l − β k = 0 ) / k 2 (\beta_k^{\mathrm{numerical}}-\beta_{k=0})/k^2 ( β k numerical − β k = 0 ) / k 2 (with β k = 0 \beta_{k=0} β k = 0 given by eq. (2) with N S t o k e s s e c t o r s = n + 2 N_{\mathrm{Stokes\ sectors}}=n+2 N Stokes sectors = n + 2 ) agrees with c 1 ( n ) c_1(n) c 1 ( n ) to an accuracy of O ( 10 − 5 ) O(10^{-5}) O ( 1 0 − 5 ) in the region k ≲ 0.01 k\lesssim0.01 k ≲ 0.01 . It is noteworthy that the topological information of N S t o k e s s e c t o r s = n + 2 N_{\mathrm{Stokes\ sectors}}=n+2 N Stokes sectors = n + 2 is ubiquitous in eq. (167) as well. If we restore the physical dispersion relationship, ω k 2 ( η ) = C ( k ) + A η n \omega_k^2(\eta)=C(k)+A\eta^n ω k 2 ( η ) = C ( k ) + A η n , the Bogoliubov coefficient contains factors of A A A , as will be displayed explicitly later in eqs. (173) and (170) . In the next subsection, we will see how c 1 ( n ) c_1(n) c 1 ( n ) controls the dominant portion of the particles produced for the class of models of ω k 2 ( η ) = k 2 + A η n \omega_k^2(\eta)=k^2+A\eta^n ω k 2 ( η ) = k 2 + A η n .
Semi-analytic approximate number density for physical applications
In this section, we combine eq. (167) and a numerically motivated ansatz to obtain a closed form approximation to the physical number density of particles produced in the dispersion relationship scenario of eq. (4) for n ≥ 4 n\geq4 n ≥ 4 and
C ( k ) = k 2 . (168) C(k)=k^2. \tag*{(168)} C ( k ) = k 2 . (168)
Focusing on C ( k ) = k 2 C(k)=k^2 C ( k ) = k 2 instead of C ( k ) = k 2 − Λ I R 2 C(k)=k^2-\Lambda_{\mathrm{IR}}^2 C ( k ) = k 2 − Λ IR 2 allows us to demonstrate the impact of the c 1 ( n ) c_1(n) c 1 ( n ) computation most strongly since without the c 1 ( n ) k 2 c_1(n)k^2 c 1 ( n ) k 2 term, the rigorous validity of the β k \beta_k β k coefficient is at a vanishing phase space point k = 0 k=0 k = 0 for C ( k ) = k 2 C(k)=k^2 C ( k ) = k 2 .
As implied by the scaling in eq. (31) , we can restore the (under)bar notation and write the physical number density of particles produced as
n χ = 1 a 3 2 π 2 ∫ k m ∞ d k k 2 ∣ β − k ˉ ( k ) ∣ 2 (169) n_\chi=\frac{1}{a^3 2\pi^2}\int_{k_m}^{\infty} dk\,k^2\left|\beta_{-\bar{k}(k)}\right|^2 \tag*{(169)} n χ = a 3 2 π 2 1 ∫ k m ∞ d k k 2 β − k ˉ ( k ) 2 (169)
where k m k_m k m is the IR cutoff on particle modes satisfying C ( k m ) = 0 C(k_m)=0 C ( k m ) = 0 (i.e. with eq. (168) , k m = 0 k_m=0 k m = 0 ) and
k ˉ ( k ) = C ( k ) A 2 n + 2 (170) \bar{k}(k)=\sqrt{\frac{C(k)}{A^{\frac{2}{n+2}}}} \tag*{(170)} k ˉ ( k ) = A n + 2 2 C ( k ) (170)
which will make the c 1 ( n ) k ˉ 2 c_1(n)\bar{k}^2 c 1 ( n ) k ˉ 2 in β k ˉ ( k ) \beta_{\bar{k}(k)} β k ˉ ( k ) non-perturbative in A A A . For any finite k max k_{\max} k m a x defining an interval [ 0 , k max ] [0,k_{\max}] [ 0 , k m a x ] and ∣ β k ˉ ( k ) p a r t ∣ |\beta^{\mathrm{part}}_{\bar{k}(k)}| ∣ β k ˉ ( k ) part ∣ definition in which ∣ β p a r t ∣ ≲ ∣ β ∣ |\beta^{\mathrm{part}}| \lesssim|\beta| ∣ β part ∣ ≲ ∣ β ∣ in that interval, a part of the total n χ n_\chi n χ can be written to reside in
n χ p a r t ≡ 1 a 3 2 π 2 ∫ 0 k max d k k 2 ∣ β k ˉ ( k ) p a r t ∣ 2 . (171) n_\chi^{\mathrm{part}} \equiv\frac{1}{a^3 2\pi^2}\int_0^{k_{\max}} dk k^2 \left|\beta^{\mathrm{part}}_{\bar{k}(k)}\right|^2. \tag*{(171)} n χ part ≡ a 3 2 π 2 1 ∫ 0 k m a x d k k 2 β k ˉ ( k ) part 2 . (171)
As justified by numerical results of Figure 5 for n = 4 , 6 , 8 n=4,6,8 n = 4 , 6 , 8 , we can choose β k ˉ ( k ) p a r t \beta^{\mathrm{part}}_{\bar{k}(k)} β k ˉ ( k ) part as
Figure 5. Comparison of different approximations for n = 4 n=4 n = 4 , 6 6 6 , 8 8 8 cases. The “quadratic approx” model is given by eq. (173) up to k ˉ = k ˉ max b d \bar{k}=\bar{k}_{\max\ \mathrm{bd}} k ˉ = k ˉ m a x bd given in eq. (176) and underestimates the particle number density. In contrast, the Gaussian approximation of eq. (180) overestimates compared to the true value.
β k ˉ p a r t = β 0 + c 1 ( n ) k ˉ 2 (172) \beta^{\mathrm{part}}_{\bar{k}}=\beta_0+c_1(n)\bar{k}^2 \tag*{(172)} β k ˉ part = β 0 + c 1 ( n ) k ˉ 2 (172)
= − i cot ( π n + 2 ) + i ∣ c 1 ( n ) ∣ C ( k ) A 2 n + 2 (173) =-i\cot\left(\frac{\pi}{n+2}\right)+i|c_1(n)|\frac{C(k)}{A^{\frac{2}{n+2}}} \tag*{(173)} = − i cot ( n + 2 π ) + i ∣ c 1 ( n ) ∣ A n + 2 2 C ( k ) (173)
in the k ˉ \bar{k} k ˉ range [ 0 , k ˉ ( k max ) ] [0,\bar{k}(k_{\max})] [ 0 , k ˉ ( k m a x )] to obtain
n χ p a r t a 3 2 π 2 = cot 2 ( π n + 2 ) [ k max 3 3 ] − 2 ∣ c 1 ( n ) ∣ A 2 n + 2 cot ( π n + 2 ) [ k max 5 5 ] + ∣ c 1 ( n ) ∣ 2 A 4 n + 2 [ k max 7 7 ] . (174) n_\chi^{\mathrm{part}}a^3 2\pi^2=\cot^2\left(\frac{\pi}{n+2}\right)\left[\frac{k_{\max}^3}{3}\right]-2\frac{|c_1(n)|}{A^{\frac{2}{n+2}}}\cot\left(\frac{\pi}{n+2}\right)\left[\frac{k_{\max}^5}{5}\right]+\frac{|c_1(n)|^2}{A^{\frac{4}{n+2}}}\left[\frac{k_{\max}^7}{7}\right]. \tag*{(174)} n χ part a 3 2 π 2 = cot 2 ( n + 2 π ) [ 3 k m a x 3 ] − 2 A n + 2 2 ∣ c 1 ( n ) ∣ cot ( n + 2 π ) [ 5 k m a x 5 ] + A n + 2 4 ∣ c 1 ( n ) ∣ 2 [ 7 k m a x 7 ] . (174)
for a judiciously chosen k max k_{\max} k m a x that we will make explicit next. Note the nontrivial A A A dependence of the c 1 c_1 c 1 dependent piece signaling the non-topological nature of these contributions. Although eq. (173) is nonperturbative in A A A through k ˉ \bar{k} k ˉ , that nonperturbativity cannot be used to make β k ˉ p a r t \beta^{\mathrm{part}}_{\bar{k}} β k ˉ part large because we are expanding in small k ˉ \bar{k} k ˉ . On the other hand the cot ( π / ( n + 2 ) ) \cot(\pi/(n+2)) cot ( π / ( n + 2 )) factor (i.e. the topological factor) can be made arbitrarily large for large n n n .
Let’s choose k max k_{\max} k m a x for the “quadratic approximation” choice of eq. (173) . Because of eq. (170) , the upper limit of the β k ˉ \beta_{\bar{k}} β k ˉ index appearing in eq. (171) is
k ˉ max = k max A − 1 n + 2 (175) \bar{k}_{\max}=k_{\max}A^{-\frac{1}{n+2}} \tag*{(175)} k ˉ m a x = k m a x A − n + 2 1 (175)
and not k ˉ max = k max \bar{k}_{\max}=k_{\max} k ˉ m a x = k m a x . The quadratic behavior makes β k ˉ p a r t \beta^{\mathrm{part}}_{\bar{k}} β k ˉ part (eq. (173) ) vanish at finite k ˉ \bar{k} k ˉ while explicit numerical computations do not indicate this behavior (see Figure 5 ). That discrepancy is obviously expected from the fact that we are computing a small k ˉ 2 \bar{k}^2 k ˉ 2 expansion in eq. (173) . Note that one can analytically compute where β k ˉ p a r t \beta^{\mathrm{part}}_{\bar{k}} β k ˉ part vanishes (see eq. (173) ) as
cot ( π 6 ) ∣ c 1 ( 4 ) ∣ ≈ 0.8 < k ˉ max b d = cot ( π n + 2 ) ∣ c 1 ( n ) ∣ < 3 2 ≈ 1.2 (176) \sqrt{\frac{\cot\left(\frac{\pi}{6}\right)}{\lvert c_1(4)\rvert}} \approx0.8 < \bar{k}_{\max\ \mathrm{bd}} = \sqrt{\frac{\cot\left(\frac{\pi}{n+2}\right)}{\lvert c_1(n)\rvert}} < \sqrt{\frac{3}{2}} \approx1.2 \tag*{(176)} ∣ c 1 ( 4 )∣ cot ( 6 π ) ≈ 0.8 < k ˉ m a x bd = ∣ c 1 ( n )∣ cot ( n + 2 π ) < 2 3 ≈ 1.2 (176)
between n = 4 n=4 n = 4 and n = ∞ n=\infty n = ∞ . Since the particle production spectrum is expected to be dominantly one bump (attested by numerical computations), we can naturally set k ˉ max = k ˉ max b d \bar{k}_{\max}=\bar{k}_{\max\ \mathrm{bd}} k ˉ m a x = k ˉ m a x bd , and find using eq. (174)
n χ p a r t = A 3 n + 2 a 3 4 cot 7 / 2 ( π n + 2 ) 105 π 2 ∣ c 1 ( n ) ∣ 3 / 2 (177) n^{\mathrm{part}}_{\chi}=\frac{A^{\frac{3}{n+2}}}{a^3}\frac{4\cot^{7/2}\left(\frac{\pi}{n+2}\right)}{105\pi^2\lvert c_1(n)\rvert^{3/2}} \tag*{(177)} n χ part = a 3 A n + 2 3 105 π 2 ∣ c 1 ( n ) ∣ 3/2 4 cot 7/2 ( n + 2 π ) (177)
where the non-perturbative nature in A A A disappeared due to k max ∝ A 1 / ( n + 2 ) k_{\max}\propto A^{1/(n+2)} k m a x ∝ A 1/ ( n + 2 ) . As one can see in Figure 5 , n χ p a r t n^{\mathrm{part}}_{\chi} n χ part underestimates the true spectrum by design. The integrated spectrum numerically for n = 4 , 6 , 8 n=4,6,8 n = 4 , 6 , 8 is
n χ s e m i - a n a l ≈ n χ p a r t × { 1.6 n = 4 1.5 n = 6 1.5 n = 8 (178) n^{\mathrm{semi\text{-}anal}}_{\chi}\approx n^{\mathrm{part}}_{\chi}\times
\begin{cases}
1.6 & n=4\\
1.5 & n=6\\
1.5 & n=8
\end{cases} \tag*{(178)} n χ semi - anal ≈ n χ part × ⎩ ⎨ ⎧ 1.6 1.5 1.5 n = 4 n = 6 n = 8 (178)
which is remarkably constant. Hence, we estimate
n χ s e m i - a n a l ≈ A 3 n + 2 a 3 2 cot 7 / 2 ( π n + 2 ) 35 π 2 ∣ c 1 ( n ) ∣ 3 / 2 (179) n^{\mathrm{semi\text{-}anal}}_{\chi}\approx\frac{A^{\frac{3}{n+2}}}{a^3}\frac{2\cot^{7/2}\left(\frac{\pi}{n+2}\right)}{35\pi^2\lvert c_1(n)\rvert^{3/2}} \tag*{(179)} n χ semi - anal ≈ a 3 A n + 2 3 35 π 2 ∣ c 1 ( n ) ∣ 3/2 2 cot 7/2 ( n + 2 π ) (179)
where c 1 ( n ) c_1(n) c 1 ( n ) is given by eq. (167) .
As another approach to use c 1 ( n ) c_1(n) c 1 ( n ) to obtain an approximate formula for n χ n_{\chi} n χ , note we can appeal to the oscillatory nature of the amplitude and stationary phase approximation intuition to consider the ansatz
∣ β k ˉ G a u s s i a n ∣ ≡ cot ( π n + 2 ) exp ( − ∣ c 1 ( n ) ∣ cot ( π n + 2 ) k ˉ 2 ) (180) \lvert\beta^{\mathrm{Gaussian}}_{\bar{k}}\rvert\equiv\cot\left(\frac{\pi}{n+2}\right)\exp\left(-\frac{\lvert c_1(n)\rvert}{\cot\left(\frac{\pi}{n+2}\right)}\bar{k}^2\right) \tag*{(180)} ∣ β k ˉ Gaussian ∣ ≡ cot ( n + 2 π ) exp ( − cot ( n + 2 π ) ∣ c 1 ( n )∣ k ˉ 2 ) (180)
(matching eq. (173) to leading k ˉ 2 \bar{k}^2 k ˉ 2 expansion) which can be numerically checked to give an overestimate approximation for different values of n n n (see Figure 5 ). This will lead to the number density
( n χ G a u s s i a n ) Λ I R = 0 = A 3 n + 2 a 3 cot 7 / 2 ( π n + 2 ) 16 2 π 3 / 2 ∣ c 1 ( n ) ∣ 3 / 2 (181) \left(n^{\mathrm{Gaussian}}_{\chi}\right)_{\Lambda_{\mathrm{IR}}=0}=\frac{A^{\frac{3}{n+2}}}{a^3}\frac{\cot^{7/2}\left(\frac{\pi}{n+2}\right)}{16\sqrt{2}\pi^{3/2}\lvert c_1(n)\rvert^{3/2}} \tag*{(181)} ( n χ Gaussian ) Λ IR = 0 = a 3 A n + 2 3 16 2 π 3/2 ∣ c 1 ( n ) ∣ 3/2 cot 7/2 ( n + 2 π ) (181)
which is an overestimate as seen in Figure 5 . What is reassuring is that the power dependences of n n n show robustness since the dependences in eqs. (179) and (181) agree. As an aside, we can express A A A as
A = a i n + 2 [ m e f f 2 ( η ) ] n + 2 2 n n ( 1 a i ∂ η m e f f 2 ( η ) [ m e f f 2 ( η ) ] 3 / 2 ) n (182) A=\frac{a_i^{n+2}\left[m_{\mathrm{eff}}^2(\eta)\right]^{\frac{n+2}{2}}}{n^n}\left(\frac{\frac{1}{a_i}\partial_\eta m_{\mathrm{eff}}^2(\eta)}{\left[m_{\mathrm{eff}}^2(\eta)\right]^{3/2}}\right)^n \tag*{(182)} A = n n a i n + 2 [ m eff 2 ( η ) ] 2 n + 2 ( [ m eff 2 ( η ) ] 3/2 a i 1 ∂ η m eff 2 ( η ) ) n (182)
where m e f f 2 ( η ) a i 2 ≡ A η n m_{\mathrm{eff}}^2(\eta)a_i^2 \equiv A\eta^n m eff 2 ( η ) a i 2 ≡ A η n (where a i a_i a i is a fiducial scale factor) to write the number density of particles produced more physically suggestively as
n χ s e m i - a n a l ≈ a i 3 a 3 ( η ) 2 cot 7 / 2 ( π n + 2 ) 35 π 2 ∣ c 1 ( n ) ∣ 3 / 2 [ m e f f 2 ( η ) ] 3 / 2 n 3 n n + 2 ( 1 a i ∂ η m e f f 2 ( η ) [ m e f f 2 ( η ) ] 3 / 2 ) 3 n n + 2 . (183) n_\chi^{\mathrm{semi\text{-}anal}} \approx\frac{a_i^3}{a^3(\eta)}\frac{2\cot^{7/2}\left(\frac{\pi}{n+2}\right)}{35\pi^2\lvert c_1(n)\rvert^{3/2}}\frac{\left[m_{\mathrm{eff}}^2(\eta)\right]^{3/2}}{n^{\frac{3n}{n+2}}}\left(\frac{\frac{1}{a_i}\partial_\eta m_{\mathrm{eff}}^2(\eta)}{\left[m_{\mathrm{eff}}^2(\eta)\right]^{3/2}}\right)^{\frac{3n}{n+2}} . \tag*{(183)} n χ semi - anal ≈ a 3 ( η ) a i 3 35 π 2 ∣ c 1 ( n ) ∣ 3/2 2 cot 7/2 ( n + 2 π ) n n + 2 3 n [ m eff 2 ( η ) ] 3/2 ( [ m eff 2 ( η ) ] 3/2 a i 1 ∂ η m eff 2 ( η ) ) n + 2 3 n . (183)
Eq. (183) shows that the particle production is a direct consequence of time-translation symmetry breaking explicitly. What is still counter-intuitive in this expression is that it looks like there is a nontrivial time-dependence through the time-dependent effective mass squared even though there is not. This is a non-generic characteristic of the model defined by eq. (4) . Another interesting item to note is that the topological factor of cot ( π / N S t o k e s s e c t o r s ) = cot ( π / ( n + 2 ) ) \cot(\pi/N_{\mathrm{Stokes\ sectors}})=\cot(\pi/(n+2)) cot ( π / N Stokes sectors ) = cot ( π / ( n + 2 )) survives here even though that naively governs only the k = 0 k=0 k = 0 part which does not have any phase space:
cot 7 / 2 ( π n + 2 ) ∣ c 1 ( n ) ∣ 3 / 2 = ( cot ( π n + 2 ) ∣ c 1 ( n ) ∣ ) 3 / 2 cot 2 ( π n + 2 ) (184) \frac{\cot^{7/2}\left(\frac{\pi}{n+2}\right)}{\lvert c_1(n)\rvert^{3/2}}=\left(\frac{\cot\left(\frac{\pi}{n+2}\right)}{\lvert c_1(n)\rvert}\right)^{3/2}\cot^2\left(\frac{\pi}{n+2}\right) \tag*{(184)} ∣ c 1 ( n ) ∣ 3/2 cot 7/2 ( n + 2 π ) = ( ∣ c 1 ( n )∣ cot ( n + 2 π ) ) 3/2 cot 2 ( n + 2 π ) (184)
where we noted in eq. (176) that cot ( π / ( n + 2 ) ) / ∣ c 1 ( n ) ∣ \cot(\pi/(n+2))/\lvert c_1(n)\rvert cot ( π / ( n + 2 )) / ∣ c 1 ( n )∣ is insensitive to n n n for large n n n .
Conclusions
By extending the exponentially improved asymptotic expansion techniques of [16 , 18 ] and discrete symmetries that involve the complexification of the momentum related variable C ( k ) C(k) C ( k ) in addition to the conformal time η \eta η , we have in eq. (173) computed the leading soft momentum contribution to the particle production amplitude in the situation with a dispersion relationship in conformal time coordinates of ω k 2 ( η ) = C ( k ) + A η n \omega_k^2(\eta)=C(k)+A\eta^n ω k 2 ( η ) = C ( k ) + A η n ∈ \in ∈ positive even as
β k = − i cot π n + 2 + i ∣ c 1 ( n ) ∣ C ( k ) A 2 n + 2 + O ( C 2 ) (185) \beta_k=-i\cot\frac{\pi}{n+2}+i\lvert c_1(n)\rvert\frac{C(k)}{A^{\frac{2}{n+2}}}+O(C^2) \tag*{(185)} β k = − i cot n + 2 π + i ∣ c 1 ( n )∣ A n + 2 2 C ( k ) + O ( C 2 ) (185)
where c 1 ( n ) c_1(n) c 1 ( n ) is given by eq. (167) and is purely imaginary. The fact that this amplitude has no non-analyticity in C C C is noteworthy given that intermediate Schroedinger phase results have non-analytic C C C dependence as seen in eqs. (161) and (I.7). Indeed, the survival of non-analytic phase would be inconsistent with the time-reversal symmetry in the system. Note also that this amplitude is non-perturbative in the coupling A A A . Combining this analytic result with numerics, we established a semi-analytic estimate of n χ n_\chi n χ given by eq. (179) where the non-perturbative signal of A − 2 / ( n + 2 ) A^{-2/(n+2)} A − 2/ ( n + 2 ) from β k \beta_k β k disappeared due to the phase space volume having powers of A 1 / ( n + 2 ) A^{1/(n+2)} A 1/ ( n + 2 ) . Nonetheless, unlike in standard saddle-point approximations where β k \beta_k β k and the number density are exponentially suppressed, the amplitude as well as the resulting number density here has no exponential suppression as evinced by cot [ π / ( n + 2 ) ] = cot [ π / N S t o k e s s e c t o r s ] \cot[\pi/(n+2)]=\cot[\pi/N_{\mathrm{Stokes\ sectors}}] cot [ π / ( n + 2 )] = cot [ π / N Stokes sectors ] which can be made arbitrarily large for large n n n .
There are several obvious future directions related to this paper. In addition to the dark matter scenario given in [9 ] , it would be useful to apply eqs. (185) and (179) to other physical scenarios. For example, there may be enhanced symmetry point type of isocurvature scenarios with nonrenormalizable operators where this formula would naturally apply. Another natural application of this paper would be to apply the results to a cosmologically viable dark matter or isocurvature scenario with C ( k m ) = 0 C(k_m)=0 C ( k m ) = 0 for k m > 0 k_m>0 k m > 0 : i.e. scenarios with tachyons. On a more technical side, there may be ways to combine the present techniques with other techniques such as the saddle-point approximation or other Stokes computational techniques to derive a less numerically reliant large k k k result leading to the integrated number density result presumably approximated by eq. (179) . One may also be able to generalize the current results to higher spins. It would also be interesting to see how one can apply the algorithmically simplified Stokes phenomena exposing exponentially improved asymptotic expansion method presented in this paper to other time-dependent cosmological scenarios.
Acknowledgments
D.J.H.C. acknowledges the support by the U.S. Department of Energy, Office of Science, Office of High Energy Physics under Award Number DE-SC0017647. N.S. and J.Z. acknowledge partial support from the Ray MacDonald Fund and Wisconsin Alumni Research Fund at the University of Wisconsin-Madison.
Analytically continuing the time coordinate
In this appendix, we provide further details on analytically continuing the time coordinate η \eta η in the physical WKB basis f ± p h y ( k , η ) f_{\pm}^{\mathrm{phy}}(k,\eta) f ± phy ( k , η ) to obtain the fictitious WKB basis f ± ( k , z ) = exp ( ± i θ ( k , z ) ) / 2 ω k ( z ) f_{\pm}(k,z)=\exp(\pm i\theta(k,z))/\sqrt{2\omega_k(z)} f ± ( k , z ) = exp ( ± i θ ( k , z )) / 2 ω k ( z ) . In particular, given the branch cuts specified in section 2.2 (see above eq. (46) and Figure 2 ), we will express the complexified dispersion relationship ω k ( z ) \omega_k(z) ω k ( z ) in terms of a sign factor determined by the topology of branch cuts and a Taylor series globally defined in the asymptotic region. This yields expression (A.8) that is used to derive the rotational symmetries in section 3. Based on the complexified ω k ( z ) \omega_k(z) ω k ( z ) , we will then specify the branch structure for ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) , thereby completing the analytic continuation of the time coordinate in the WKB basis.
Consider ω k ( z ) = k 2 + z n \omega_k(z)=\sqrt{k^2+z^n} ω k ( z ) = k 2 + z n that is globally defined as a function of complex number z z z (see Figure 3 ) and analytic except along the branch cuts which we choose to be coincident with the radially emanating anti-Stokes boundaries. Because of the branch cuts, this original representation of ω k ( z ) \omega_k(z) ω k ( z ) is naturally discontinuous across each of the branch cuts on the first Riemann sheet even though it is continuous within the disk of radius ∣ ψ j ∣ |\psi_j| ∣ ψ j ∣ (the modulus of the branch point, recall that ψ j = k 2 / n e ( 2 j − 1 ) i π / n \psi_j=k^{2/n}e^{(2j-1)i\pi/n} ψ j = k 2/ n e ( 2 j − 1 ) iπ / n ): a region which we call inner disk D D D . On the other hand, because ω k ( z ) \omega_k(z) ω k ( z ) naturally accesses the second Riemann sheet under rotational algebraic manipulation ω k ( z e i θ ) \omega_k(ze^{i\theta}) ω k ( z e i θ ) in an unintended manner for ∣ z ∣ > ∣ ψ j ∣ |z|>|\psi_j| ∣ z ∣ > ∣ ψ j ∣ (for example in the asymptotic annulus (AA)), working with only the original ω k ( z ) \omega_k(z) ω k ( z ) representation leads to an opaque construction of the rotational symmetry representation that we build in section 3.
To remedy this problem and make the symmetry representation construction more transparent, we use a function ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) which cannot access the second Riemann sheet in z z z under rotation and represent the ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) staying strictly on the first Riemann sheet as
ω ‾ k ( z ) = { k 2 + z n z ∈ D s ( z ) ω k ( z ) ∣ z ∣ > ∣ ψ j ∣ (186) \underline{\omega}_k(z) =
\begin{cases}
\sqrt{k^2+z^n} & z \in D \\
s(z)\omega_k(z) & |z| > |\psi_j|
\end{cases} \tag*{(186)} ω k ( z ) = { k 2 + z n s ( z ) ω k ( z ) z ∈ D ∣ z ∣ > ∣ ψ j ∣ (186)
where ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) is defined as the Taylor expansion
ω ‾ k ( z ) = z n / 2 ∑ m = 0 ∞ ( 1 / 2 m ) ( k 2 z n ) m (187) \underline{\omega}_k(z) = z^{n/2}\sum_{m=0}^{\infty}\binom{1/2}{m}\left(\frac{k^2}{z^n}\right)^m \tag*{(187)} ω k ( z ) = z n /2 m = 0 ∑ ∞ ( m 1/2 ) ( z n k 2 ) m (187)
which is analytic in z z z (since n n n is even) and converges for ∣ z ∣ ≥ ∣ ψ j ∣ |z| \ge|\psi_j| ∣ z ∣ ≥ ∣ ψ j ∣ but not for ∣ z ∣ < ∣ ψ j ∣ |z| < |\psi_j| ∣ z ∣ < ∣ ψ j ∣ . Since the disk D D D defined by ∣ z ∣ ≤ ∣ ψ j ∣ |z| \le|\psi_j| ∣ z ∣ ≤ ∣ ψ j ∣ for j ∈ { 1 , … , n } j \in\{1,\ldots,n\} j ∈ { 1 , … , n } is therefore the natural boundary for ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) , we define branch sectors B j \mathcal{B}_j B j as the region bounded by two branch cuts and outside of the disk D D D . We will conventionally put the B j = 1 \mathcal{B}_{j=1} B j = 1 to include the positive real axis and go counter-clockwise in assigning the indices. Because ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) has no branch cuts in the AA nor for any ∣ z ∣ > ∣ ψ j ∣ |z| > |\psi_j| ∣ z ∣ > ∣ ψ j ∣ (in contrast with ω k ( z ) \omega_k(z) ω k ( z ) ), its rotational transformation is manifestly trivial. All the nontrivialities of the rotation are in the discrete branch cut information which is now separated in s ( z ) s(z) s ( z ) . We will determine s ( z ) s(z) s ( z ) below by matching to the properties of ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) .
The original function ω k ( z ) \omega_k(z) ω k ( z ) is analytic on B 1 \mathcal{B}_1 B 1 that includes positive real axis just like ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) and their values coincide by construction. Therefore, ω k ( z ) \omega_k(z) ω k ( z ) is isomorphic to ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) in this region such that s ( z ) s(z) s ( z ) of eq. (A.1) takes the value of unity in this region. To find the function s ( z ) s(z) s ( z ) in the other B j \mathcal{B}_j B j regions, note that since
ω k 2 ( z ) = z n ( 1 + k 2 z n ) = ω ‾ k 2 ( z ) (188) \omega_k^2(z)=z^n\left(1+\frac{k^2}{z^n}\right)=\underline{\omega}_k^2(z) \tag*{(188)} ω k 2 ( z ) = z n ( 1 + z n k 2 ) = ω k 2 ( z ) (188)
is analytic in all ∣ z ∣ > ∣ ψ j ∣ |z| > |\psi_j| ∣ z ∣ > ∣ ψ j ∣ , we find the identity
ω k ( z ) = ± ω ‾ k ( z ) (189) \omega_k(z)=\pm\underline{\omega}_k(z) \tag*{(189)} ω k ( z ) = ± ω k ( z ) (189)
in any given branch sector B j \mathcal{B}_j B j .22 This says that s ( z ) s(z) s ( z ) on any given B j \mathcal{B}_j B j is a constant sign that only depends on j j j since both ω k ( z ) \omega_k(z) ω k ( z ) and ω ‾ k ( z ) \underline{\omega}_k(z) ω k ( z ) are analytic in B j \mathcal{B}_j B j . To find the dependence on j j j , consider ω k ( z ) \omega_k(z) ω k ( z ) on the two adjacent points z + ∈ B j + 1 z_+ \in\mathcal{B}_{j+1} z + ∈ B j + 1 and z − ∈ B j z_- \in\mathcal{B}_j z − ∈ B j on either side of a branch cut to establish a relative minus sign:
ω k ( z + ) = − ω k ( z − ) (190) \omega_k(z_+)=-\omega_k(z_-) \tag*{(190)} ω k ( z + ) = − ω k ( z − ) (190)
which says that s ( z ) s(z) s ( z ) flips signs every time j j j in B j \mathcal{B}_j B j increments by unity. Hence, we have shown
ω k ( z ) = { k 2 + z n z ∈ D ( − 1 ) q − 1 ω ‾ k ( z ) z ∈ B q (191) \omega_k(z)=
\begin{cases}
\sqrt{k^2+z^n} & z \in D \\
(-1)^{q-1}\underline{\omega}_k(z) & z \in\mathcal{B}_q
\end{cases} \tag*{(191)} ω k ( z ) = { k 2 + z n ( − 1 ) q − 1 ω k ( z ) z ∈ D z ∈ B q (191)
or equivalently s ( z ) = ( − 1 ) q − 1 s(z)=(-1)^{q-1} s ( z ) = ( − 1 ) q − 1 for any z ∈ B q z\in\mathcal{B}_q z ∈ B q . By construction, we see now that ω k ( z e i θ ) \omega_k(ze^{i\theta}) ω k ( z e i θ ) cannot access the second Riemann sheet in z z z in the representation of the right hand side of equation (A.6). Note that if we want to be more formulaically explicit regarding the Stokes sector embedding, then we can write S j ⊂ B I ( j ) \mathcal{S}_j\subset\mathcal{B}_{\mathcal{I}(j)} S j ⊂ B I ( j ) with
I ( j ) = { j if j ∈ { 1 , … , n / 2 + 1 } [ ( j − 2 ) m o d n ] + 1 if j ∈ { n / 2 + 2 , … , n + 2 } (192) \mathcal{I}(j)=
\begin{cases}
j & \text{if }j\in\{1,\ldots,n/2+1\}\\
[(j-2)\bmod n]+1 & \text{if }j\in\{n/2+2,\ldots,n+2\}
\end{cases} \tag*{(192)} I ( j ) = { j [( j − 2 ) mod n ] + 1 if j ∈ { 1 , … , n /2 + 1 } if j ∈ { n /2 + 2 , … , n + 2 } (192)
Even more explicit is the form in terms of Stokes sector label instead:
ω k ( z ) = { k 2 + z n z ∈ D s j ω k ( z ) z ∈ S j E (193) \omega_k(z)=
\begin{cases}
\sqrt{k^2+z^n} & z\in D\\
s_j\omega_k(z) & z\in\mathcal{S}_j^E
\end{cases} \tag*{(193)} ω k ( z ) = { k 2 + z n s j ω k ( z ) z ∈ D z ∈ S j E (193)
where
s j = { ( − 1 ) j − 1 1 ≤ j ≤ n / 2 + 1 ( − 1 ) j n / 2 + 2 ≤ j ≤ n + 2 (194) s_j=
\begin{cases}
(-1)^{j-1} & 1\le j\le n/2+1\\
(-1)^j & n/2+2\le j\le n+2
\end{cases} \tag*{(194)} s j = { ( − 1 ) j − 1 ( − 1 ) j 1 ≤ j ≤ n /2 + 1 n /2 + 2 ≤ j ≤ n + 2 (194)
and where the superscript “E” in S j E \mathcal{S}_j^E S j E corresponds to the extension of the Stokes sector to cover the region between the Stokes sector and the inner disk. Since we will be rotating between Stokes sectors in § 3 , it will be more convenient to use s j s_j s j which is a function of S j E \mathcal{S}_j^E S j E indices j j j and not the branch sector indices B j \mathcal{B}_j B j .
Next, we analytically continue the time coordinate in the square root ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) appearing in the denominator of the fictitious WKB basis f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) . Since we want f ± ( k , z ) f_\pm(k,z) f ± ( k , z ) to match the physical WKB basis f ± p h y ( k , η ) f_\pm^{\mathrm{phy}}(k,\eta) f ± phy ( k , η ) on the positive real axis, we require ω k ( z ) > 0 \sqrt{\omega_k(z)}>0 ω k ( z ) > 0 for z ∈ R > 0 z\in\mathbb{R}_{>0} z ∈ R > 0 . The values of ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) elsewhere are then fixed by continuity. Inside the inner disk D D D , we have
ω k ( z ) = + ( k 2 + z n ) 1 / 4 , z ∈ D . (195) \sqrt{\omega_k(z)}=+(k^2+z^n)^{1/4},\qquad z\in D. \tag*{(195)} ω k ( z ) = + ( k 2 + z n ) 1/4 , z ∈ D . (195)
Outside the D D D , using eq. (A.6) we get
ω k ( z ) = ( − 1 ) q − 1 ω k ( z ) = ( − 1 ) q − 1 ω k ( z ) , z ∈ B q (196) \sqrt{\omega_k(z)}=\sqrt{(-1)^{q-1}\omega_k(z)}=\sqrt{(-1)^{q-1}}\sqrt{\omega_k(z)},\qquad z\in\mathcal{B}_q \tag*{(196)} ω k ( z ) = ( − 1 ) q − 1 ω k ( z ) = ( − 1 ) q − 1 ω k ( z ) , z ∈ B q (196)
where the square root branches of ( − 1 ) q − 1 \sqrt{(-1)^{q-1}} ( − 1 ) q − 1 and ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) are fixed by continuity23
( − 1 ) q − 1 = e − i π 2 ( q − 1 ) (197) \sqrt{(-1)^{q-1}}=e^{-\frac{i\pi}{2}(q-1)} \tag*{(197)} ( − 1 ) q − 1 = e − 2 iπ ( q − 1 ) (197)
ω k ( z ) = ∣ z ∣ n 4 e i n 4 arg z ∑ m = 0 ∞ ( 1 / 4 m ) ( k 2 z n ) m , ∣ z ∣ > k 2 / n . (198) \sqrt{\omega_k(z)}=|z|^{\frac{n}{4}}e^{\frac{in}{4}\arg z}\sum_{m=0}^{\infty}\binom{1/4}{m}\left(\frac{k^2}{z^n}\right)^m,\qquad|z|>k^{2/n}. \tag*{(198)} ω k ( z ) = ∣ z ∣ 4 n e 4 in a r g z m = 0 ∑ ∞ ( m 1/4 ) ( z n k 2 ) m , ∣ z ∣ > k 2/ n . (198)
Thus, the analytic continuation of ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) is
ω k ( z ) = { + ( k 2 + z n ) 1 / 4 z ∈ D e − i π 2 ( q − 1 ) ω k ( z ) z ∈ B q (199) \sqrt{\omega_k(z)}=
\begin{cases}
+(k^2+z^n)^{1/4} & z\in D\\
e^{-\frac{i\pi}{2}(q-1)}\sqrt{\omega_k(z)} & z\in\mathcal{B}_q
\end{cases} \tag*{(199)} ω k ( z ) = { + ( k 2 + z n ) 1/4 e − 2 iπ ( q − 1 ) ω k ( z ) z ∈ D z ∈ B q (199)
with ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) specified by eq. (A.13). Consequently, the analytical continuation ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) matches the ω k ( η ) \sqrt{\omega_k(\eta)} ω k ( η ) in the physical WKB basis. To see this, note they match on the positive real axis by construction. On the negative real axis, they match in the inner disk D D D because of eq. (A.10), and also match for z ∈ R < − k 2 / n z \in\mathbb{R}_{<-k^{2/n}} z ∈ R < − k 2/ n because
ω k ( z ) ∣ z ∈ R < − k 2 / n = e − i π N 2 ∣ z ∣ N 2 e i π N 2 ∑ m = 0 ∞ ( 1 / 4 m ) ( k 2 z n ) m = ∣ z ∣ N 2 ∑ m = 0 ∞ ( 1 / 4 m ) ( k 2 z n ) m > 0. (200) \left.\sqrt{\omega_k(z)}\right|_{z\in\mathbb{R}_{<-k^{2/n}}}
= e^{-\frac{i\pi N}{2}} |z|^{\frac{N}{2}} e^{\frac{i\pi N}{2}} \sum_{m=0}^{\infty} \binom{1/4}{m}\left(\frac{k^2}{z^n}\right)^m
= |z|^{\frac{N}{2}} \sum_{m=0}^{\infty} \binom{1/4}{m}\left(\frac{k^2}{z^n}\right)^m > 0 . \tag*{(200)} ω k ( z ) z ∈ R < − k 2/ n = e − 2 iπ N ∣ z ∣ 2 N e 2 iπ N m = 0 ∑ ∞ ( m 1/4 ) ( z n k 2 ) m = ∣ z ∣ 2 N m = 0 ∑ ∞ ( m 1/4 ) ( z n k 2 ) m > 0. (200)
The analytic continuations of ω k ( z ) \omega_k(z) ω k ( z ) and ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) have two further implications. First, the fictitious WKB basis functions defined by eq. (18) ,
f ± ( k , z ) = e ± i ∫ 0 z d z ′ ω k ( z ′ ) 2 ω k ( z ) , (201) f_\pm(k,z) = \frac{e^{\pm i\int_0^z dz'\,\omega_k(z')}}{\sqrt{2\omega_k(z)}} , \tag*{(201)} f ± ( k , z ) = 2 ω k ( z ) e ± i ∫ 0 z d z ′ ω k ( z ′ ) , (201)
are uniquely specified by the analytic continuations (A.6) and (A.14). This analytic continuation matches the fictitious and physical WKB bases throughout the real axis. This means the Bogoliubov β k \beta_k β k computed using the propagator expression (48) of the fictitious WKB basis is equal to that computed using the Wronskian projection (16) of the physical WKB basis. Secondly, the analytic continuation of ω k ( z ) \sqrt{\omega_k(z)} ω k ( z ) implies that it gains a ( − i ) (-i) ( − i ) factor when crossing a branch cut counterclockwise. This is used in deriving eq. (44) , the transformation of the fictitious WKB basis across branch cuts.
Complexifying the momentum variable
In this appendix, we provide a different perspective for complexifying the physical momentum variable k ∈ R > 0 k \in\mathbb{R}_{>0} k ∈ R > 0 in the WKB phase θ ( k , z ) = ∫ 0 z d z ′ ω k ( z ′ ) \theta(k,z) = \int_0^z dz'\,\omega_k(z') θ ( k , z ) = ∫ 0 z d z ′ ω k ( z ′ ) to obtain θ ( p , z ) \theta(p,z) θ ( p , z ) , where p ∈ C p \in\mathbb{C} p ∈ C denotes the complexified momentum. The complexified WKB phase θ ( p , z ) \theta(p,z) θ ( p , z ) is used to derive the rotational symmetry in § 3.1 and to isolate the k k k non-analyticities in § 4.1 . Throughout this appendix we assume n ≥ 4 n \ge4 n ≥ 4 .
Our prescription for complexification is to split θ ( k , z ) \theta(k,z) θ ( k , z ) into terms that are analytic and non-analytic in k k k . We complexify the analytic term by replacing k k k with the complex momentum p p p , while the non-analytic terms are continued by specifying the Riemann sheet in the complex p p p -plane. As a generalization to § 4.1 , for z ∈ S j z \in\mathcal{S}_j z ∈ S j , this decomposition can be achieved by splitting the WKB phase at ψ j \psi_j ψ j , the endpoint of the j j j -th branch cut on the z z z -plane, which intuitively captures the non-analytic dependence on k k k . So we write
θ ( k , z ) = ∫ 0 ψ j d z ′ ω k ( z ′ ) + ∫ ψ j z d z ′ ω k ( z ′ ) = Δ j ( k ) + ∫ ψ j z d z ′ ω k ( z ′ ) (202) \theta(k,z) = \int_0^{\psi_j} dz'\,\omega_k(z') + \int_{\psi_j}^{z} dz'\,\omega_k(z') = \Delta_j(k) + \int_{\psi_j}^{z} dz'\,\omega_k(z') \tag*{(202)} θ ( k , z ) = ∫ 0 ψ j d z ′ ω k ( z ′ ) + ∫ ψ j z d z ′ ω k ( z ′ ) = Δ j ( k ) + ∫ ψ j z d z ′ ω k ( z ′ ) (202)
where recall that Δ j ( k ) = ∫ 0 ψ j d z ω k ( z ) \Delta_j(k) = \int_0^{\psi_j} dz\,\omega_k(z) Δ j ( k ) = ∫ 0 ψ j d z ω k ( z ) denotes the phase factor, which can be computed with a change of variable
Δ j ( k ) = ∫ 0 ψ j d z k 2 + z n = e 2 j − 1 n i π k n + 2 n d n , d n ≡ Γ ( 3 2 ) Γ ( 1 + 1 n ) Γ ( 3 2 + 1 n ) , (203) \Delta_j(k) = \int_0^{\psi_j} dz\sqrt{k^2+z^n} = e^{\frac{2j-1}{n}i\pi} k^{\frac{n+2}{n}} d_n,\qquad d_n \equiv\frac{\Gamma\left(\frac{3}{2}\right)\Gamma\left(1+\frac{1}{n}\right)}{\Gamma\left(\frac{3}{2}+\frac{1}{n}\right)}, \tag*{(203)} Δ j ( k ) = ∫ 0 ψ j d z k 2 + z n = e n 2 j − 1 iπ k n n + 2 d n , d n ≡ Γ ( 2 3 + n 1 ) Γ ( 2 3 ) Γ ( 1 + n 1 ) , (203)
To further evaluate the second term in eq. (B.1), we use the Taylor series expression derived in eq. (A.8)
ω k ( z ) = s j z n / 2 ∑ m = 0 ∞ ( 1 / 2 m ) ( k 2 z n ) m , z ∈ S j E . (204) \omega_k(z)=s_j z^{n/2}\sum_{m=0}^{\infty}\binom{1/2}{m}\left(\frac{k^2}{z^n}\right)^m,\qquad z\in\mathcal{S}_j^E. \tag*{(204)} ω k ( z ) = s j z n /2 m = 0 ∑ ∞ ( m 1/2 ) ( z n k 2 ) m , z ∈ S j E . (204)
This series converges uniformly for z ∈ S j E z\in\mathcal{S}_j^E z ∈ S j E , so we can integrate it term by term to get
∫ ψ j z d z ′ ω k ( z ′ ) = s j z n / 2 + 1 n / 2 + 1 − s j A ( k , z ) + s j Ψ j ( k ) (205) \int_{\psi_j}^{z} dz'\omega_k(z')=s_j\frac{z^{n/2+1}}{n/2+1}-s_jA(k,z)+s_j\Psi_j(k) \tag*{(205)} ∫ ψ j z d z ′ ω k ( z ′ ) = s j n /2 + 1 z n /2 + 1 − s j A ( k , z ) + s j Ψ j ( k ) (205)
where
A ( k , z ) ≡ ∑ m = 1 ∞ ( 1 / 2 m ) 1 m n − ( n / 2 + 1 ) k 2 m z m n − ( n / 2 + 1 ) , (206) A(k,z)\equiv\sum_{m=1}^{\infty}\binom{1/2}{m}\frac{1}{mn-(n/2+1)}\frac{k^{2m}}{z^{mn-(n/2+1)}}, \tag*{(206)} A ( k , z ) ≡ m = 1 ∑ ∞ ( m 1/2 ) mn − ( n /2 + 1 ) 1 z mn − ( n /2 + 1 ) k 2 m , (206)
Ψ j ( k ) ≡ ∑ m = 0 ∞ ( 1 / 2 m ) 1 m n − ( n / 2 + 1 ) k 2 m ψ j m n − ( n / 2 + 1 ) . (207) \Psi_j(k)\equiv\sum_{m=0}^{\infty}\binom{1/2}{m}\frac{1}{mn-(n/2+1)}\frac{k^{2m}}{\psi_j^{mn-(n/2+1)}}. \tag*{(207)} Ψ j ( k ) ≡ m = 0 ∑ ∞ ( m 1/2 ) mn − ( n /2 + 1 ) 1 ψ j mn − ( n /2 + 1 ) k 2 m . (207)
Here we note for z ∈ S j E z\in\mathcal{S}_j^E z ∈ S j E , A ( k , z ) A(k,z) A ( k , z ) is analytic in k k k . The Ψ j ( k ) \Psi_j(k) Ψ j ( k ) term can be further simplified using ψ j = k 2 / n e i ( 2 j − 1 ) π / n \psi_j=k^{2/n}e^{i(2j-1)\pi/n} ψ j = k 2/ n e i ( 2 j − 1 ) π / n :
Ψ j ( k ) = k ( n + 2 ) / n ∑ m = 0 ∞ ( 1 / 2 m ) e − i 2 j − 1 n π ( n m − n / 2 − 1 ) n m − ( n / 2 + 1 ) . (208) \Psi_j(k)=k^{(n+2)/n}\sum_{m=0}^{\infty}\binom{1/2}{m}\frac{e^{-i\frac{2j-1}{n}\pi(nm-n/2-1)}}{nm-(n/2+1)}. \tag*{(208)} Ψ j ( k ) = k ( n + 2 ) / n m = 0 ∑ ∞ ( m 1/2 ) nm − ( n /2 + 1 ) e − i n 2 j − 1 π ( nm − n /2 − 1 ) . (208)
which implies Ψ j ( k ) \Psi_j(k) Ψ j ( k ) is proportional to k ( n + 2 ) / n k^{(n+2)/n} k ( n + 2 ) / n as the series in m m m is convergent, making Ψ j ( k ) \Psi_j(k) Ψ j ( k ) non-analytic in k k k . Combining eqs. (B.1) and (B.4) we obtain the decomposition
θ ( k , z ) = s j z n / 2 + 1 n / 2 + 1 − s j A ( k , z ) ⏟ analytic in k + Δ j ( k ) + s j Ψ j ( k ) ⏟ non-analytic in k , k ∈ R ≥ 0 , z ∈ S j E . (209) \theta(k,z)=\underbrace{s_j\frac{z^{n/2+1}}{n/2+1}-s_jA(k,z)}_{\text{analytic in }k}+\underbrace{\Delta_j(k)+s_j\Psi_j(k)}_{\text{non-analytic in }k},\qquad k\in\mathbb{R}_{\geq0},\quad z\in\mathcal{S}_j^E. \tag*{(209)} θ ( k , z ) = analytic in k s j n /2 + 1 z n /2 + 1 − s j A ( k , z ) + non-analytic in k Δ j ( k ) + s j Ψ j ( k ) , k ∈ R ≥ 0 , z ∈ S j E . (209)
The analytic term is then complexified by replacing k k k with p p p in the power series expression (B.5)
s j A ( k , z ) → s j A ( p , z ) (210) s_jA(k,z)\to s_jA(p,z) \tag*{(210)} s j A ( k , z ) → s j A ( p , z ) (210)
with
A ( p , z ) ≡ ∑ m = 1 ∞ ( 1 / 2 m ) 1 m n − ( n / 2 + 1 ) p 2 m z m n − ( n / 2 + 1 ) , p ∈ C . (211) A(p,z)\equiv\sum_{m=1}^{\infty}\binom{1/2}{m}\frac{1}{mn-(n/2+1)}\frac{p^{2m}}{z^{mn-(n/2+1)}},\qquad p\in\mathbb{C}. \tag*{(211)} A ( p , z ) ≡ m = 1 ∑ ∞ ( m 1/2 ) mn − ( n /2 + 1 ) 1 z mn − ( n /2 + 1 ) p 2 m , p ∈ C . (211)
We next consider the non-analytic terms Δ j ( k ) \Delta_j(k) Δ j ( k ) and Ψ j ( k ) \Psi_j(k) Ψ j ( k ) . Both can be expressed in terms of the branch point ψ j \psi_j ψ j . Since integration contour for Δ j ( k ) \Delta_j(k) Δ j ( k ) lies inside the inner disk D D D , we may Taylor expand ω k ( z ′ ) \omega_k(z') ω k ( z ′ ) in powers of z ′ n / k 2 z'^n/k^2 z ′ n / k 2 and integrate term by term:
Δ j ( k ) = ∫ 0 ψ j d z ′ k ∑ m = 0 ∞ ( 1 / 2 m ) ( z ′ n k 2 ) m = [ k ψ j ( k ) ] ∑ m = 0 ∞ ( 1 / 2 m ) 1 n m + 1 [ ψ j ( k ) ] n m k 2 m . (212) \begin{aligned}
\Delta_j(k)&=\int_0^{\psi_j} dz'k\sum_{m=0}^{\infty}\binom{1/2}{m}\left(\frac{z'^n}{k^2}\right)^m\\
&=[k\psi_j(k)]\sum_{m=0}^{\infty}\binom{1/2}{m}\frac{1}{nm+1}\frac{[\psi_j(k)]^{nm}}{k^{2m}}. \tag*{(212)}
\end{aligned} Δ j ( k ) = ∫ 0 ψ j d z ′ k m = 0 ∑ ∞ ( m 1/2 ) ( k 2 z ′ n ) m = [ k ψ j ( k )] m = 0 ∑ ∞ ( m 1/2 ) nm + 1 1 k 2 m [ ψ j ( k ) ] nm . (212)
On the other hand, eq. (B.6) can be rewritten as
Ψ j ( k ) = [ ψ j ( k ) ] n / 2 + 1 ∑ m = 0 ∞ ( 1 / 2 m ) 1 n m − ( n / 2 + 1 ) k 2 m [ ψ j ( k ) ] n m . (213) \Psi_j(k)=[\psi_j(k)]^{n/2+1}\sum_{m=0}^{\infty}\binom{1/2}{m}\frac{1}{nm-(n/2+1)}\frac{k^{2m}}{[\psi_j(k)]^{nm}}. \tag*{(213)} Ψ j ( k ) = [ ψ j ( k ) ] n /2 + 1 m = 0 ∑ ∞ ( m 1/2 ) nm − ( n /2 + 1 ) 1 [ ψ j ( k ) ] nm k 2 m . (213)
Since ψ j ∝ k 2 / n \psi_j\propto k^{2/n} ψ j ∝ k 2/ n implies ψ j n m ∝ k 2 m \psi_j^{nm}\propto k^{2m} ψ j nm ∝ k 2 m , the factors ψ j n m \psi_j^{nm} ψ j nm are analytic in k k k . It is therefore sufficient to obtain the complexifications Δ j ( p ) \Delta_j(p) Δ j ( p ) and Ψ j ( p ) \Psi_j(p) Ψ j ( p ) by specifying the analytic continuation of the branch point ψ j ( p ) \psi_j(p) ψ j ( p ) . By definition, ψ j ( p ) \psi_j(p) ψ j ( p ) is the complexification of the j j j -th zero of ω k ( z ) \omega_k(z) ω k ( z ) , which should satisfy p 2 + [ ψ j ( p ) ] n = 0 p^2+[\psi_j(p)]^n=0 p 2 + [ ψ j ( p ) ] n = 0 . Specifically, we choose
ψ j ( p ) = e i ( 2 j − 1 ) π / n ∣ p ∣ 2 / n e i ( 2 / n ) arg p , arg p ∈ ( − 2 π , 0 ] . (214) \psi_j(p)=e^{i(2j-1)\pi/n}|p|^{2/n}e^{i(2/n)\arg p},\qquad\arg p\in(-2\pi,0]. \tag*{(214)} ψ j ( p ) = e i ( 2 j − 1 ) π / n ∣ p ∣ 2/ n e i ( 2/ n ) a r g p , arg p ∈ ( − 2 π , 0 ] . (214)
Here we have the freedom to choose the branch cut for ψ j ( p ) \psi_j(p) ψ j ( p ) as long as the cut is not on the positive real momentum axis, then ψ j ( p = k ) = e i ( 2 j − 1 ) π / n k 2 / n \psi_j(p=k)=e^{i(2j-1)\pi/n}k^{2/n} ψ j ( p = k ) = e i ( 2 j − 1 ) π / n k 2/ n is guaranteed. Since we are interested in rotating k k k to ( γ − n / 2 ) j k (\gamma^{-n/2})^j k ( γ − n /2 ) j k with 1 ≤ j ≤ n + 1 1\leq j\leq n+1 1 ≤ j ≤ n + 1 as in § 3.1 , we stipulate arg p ∈ ( − 2 π , 0 ] \arg p\in(-2\pi,0] arg p ∈ ( − 2 π , 0 ] to fix the branch cut for ψ j ( p ) \psi_j(p) ψ j ( p ) . This in turn fixes the branch cut for θ ( p , z ) \theta(p,z) θ ( p , z ) , giving the complexified WKB phase
θ ( p , z ) = s j z n / 2 + 1 n / 2 + 1 − s j A ( p , z ) + Δ j ( p ) + s j Ψ j ( p ) , arg p ∈ ( − 2 π , 0 ] , z ∈ S j . (215) \theta(p,z)=s_j\frac{z^{n/2+1}}{n/2+1}-s_jA(p,z)+\Delta_j(p)+s_j\Psi_j(p),\qquad\arg p\in(-2\pi,0],\ z\in\mathcal{S}_j. \tag*{(215)} θ ( p , z ) = s j n /2 + 1 z n /2 + 1 − s j A ( p , z ) + Δ j ( p ) + s j Ψ j ( p ) , arg p ∈ ( − 2 π , 0 ] , z ∈ S j . (215)
with the analytic term A ( p , z ) A(p,z) A ( p , z ) given by eq. (B.10) and the non-analytic terms given by
Δ j ( p ) = ψ j ( p ) ∑ m = 0 ∞ ( 1 / 2 m ) 1 n m + 1 [ ψ j ( p ) ] n m p 2 m − 1 . (216) \Delta_j(p)=\psi_j(p)\sum_{m=0}^{\infty}\binom{1/2}{m}\frac{1}{nm+1}\frac{[\psi_j(p)]^{nm}}{p^{2m-1}}. \tag*{(216)} Δ j ( p ) = ψ j ( p ) m = 0 ∑ ∞ ( m 1/2 ) nm + 1 1 p 2 m − 1 [ ψ j ( p ) ] nm . (216)
Ψ j ( p ) = [ ψ j ( p ) ] n / 2 + 1 ∑ m = 0 ∞ ( 1 / 2 m ) 1 n m − ( n / 2 + 1 ) p 2 m [ ψ j ( p ) ] n m . (217) \Psi_j(p)=[\psi_j(p)]^{n/2+1}\sum_{m=0}^{\infty}\binom{1/2}{m}\frac{1}{nm-(n/2+1)}\frac{p^{2m}}{[\psi_j(p)]^{nm}}. \tag*{(217)} Ψ j ( p ) = [ ψ j ( p ) ] n /2 + 1 m = 0 ∑ ∞ ( m 1/2 ) nm − ( n /2 + 1 ) 1 [ ψ j ( p ) ] nm p 2 m . (217)
This is the main result of this appendix.
Strictly speaking, we should have included the p p p dependence in the Stokes sector S j \mathcal{S}_j S j , since varying arg p \arg p arg p rotates the zeros ψ j ( p ) \psi_j(p) ψ j ( p ) and hence drags the branch cuts emanating from them. Nevertheless, as shown in Figure 4 , we may choose to keep the branch-cut segments fixed in the AA under this continuation. The corresponding branch sectors, and therefore the factors s j s_j s j , may then be kept fixed. We therefore suppress the momentum dependence of the notation S j \mathcal{S}_j S j .
Relation among Stokes constants
Here we show how to obtain the relation (70) –(72) among the Stokes constants from the rotational symmetry representation on the propagators (69) and the property (74) of the phase factor Δ j ( k ) \Delta_j(k) Δ j ( k ) .
We start with the propagator U 1 ( k ) U_1(k) U 1 ( k ) across the first Stokes sector, and rotate counterclockwise to obtain all the other propagators U j ( k ) U_j(k) U j ( k ) . First, we introduce the notations
U − [ S ] ≡ ( 1 0 S 1 ) , U + [ S ] ≡ ( 1 S 0 1 ) , R [ x ] ≡ ( e 2 i x 0 0 e − 2 i x ) . (218) U_-[S]\equiv\begin{pmatrix}1&0\\S&1\end{pmatrix},\qquad U_+[S]\equiv\begin{pmatrix}1&S\\0&1\end{pmatrix},\qquad\mathcal{R}[x]\equiv\begin{pmatrix}e^{2ix}&0\\0&e^{-2ix}\end{pmatrix}. \tag*{(218)} U − [ S ] ≡ ( 1 S 0 1 ) , U + [ S ] ≡ ( 1 0 S 1 ) , R [ x ] ≡ ( e 2 i x 0 0 e − 2 i x ) . (218)
It follows immediately that
R − 1 [ x ] = R [ − x ] , R [ x ] R [ y ] = R [ x + y ] , R [ x ] U ± [ S ] R − 1 [ x ] = U ± [ e ± 4 i x S ] . (219) \mathcal{R}^{-1}[x]=\mathcal{R}[-x],\qquad\mathcal{R}[x]\mathcal{R}[y]=\mathcal{R}[x+y],\qquad\mathcal{R}[x]U_{\pm}[S]\mathcal{R}^{-1}[x]=U_{\pm}[e^{\pm4ix}S]. \tag*{(219)} R − 1 [ x ] = R [ − x ] , R [ x ] R [ y ] = R [ x + y ] , R [ x ] U ± [ S ] R − 1 [ x ] = U ± [ e ± 4 i x S ] . (219)
On the upper half plane, the rotational symmetry representation (69) reads
U − [ S j + 1 ( k ) ] = R [ Δ j ( k ) ] U − [ S j ( γ − N k ) ] R − 1 [ Δ j ( k ) ] , 1 ≤ j ≤ N , (220) U_{-}[S_{j+1}(k)]=\mathcal{R}[\Delta_j(k)]U_{-}[S_j(\gamma^{-N}k)]\mathcal{R}^{-1}[\Delta_j(k)],\qquad1\leq j\leq N, \tag*{(220)} U − [ S j + 1 ( k )] = R [ Δ j ( k )] U − [ S j ( γ − N k )] R − 1 [ Δ j ( k )] , 1 ≤ j ≤ N , (220)
Iterating the above relation and using eqs. (74) and (C.2), for 1 ≤ j ≤ N 1\leq j\leq N 1 ≤ j ≤ N we obtain
U − [ S j + 1 ( k ) ] = R [ Δ j ( k ) ] R [ Δ j − 1 ( γ − N k ) ] U − [ S j − 1 ( γ − 2 N k ) ] R − 1 [ Δ j − 1 ( γ − N k ) ] R − 1 [ Δ j ( k ) ] = R [ Δ ~ j ( k ) ] U − [ S 1 ( γ − N j k ) ] R − 1 [ Δ ~ j ( k ) ] = U − [ e − 4 i Δ ~ j ( k ) S 1 ( γ − N j k ) ] . (221) \begin{aligned}
U_{-}[S_{j+1}(k)]&=\mathcal{R}[\Delta_j(k)]\mathcal{R}[\Delta_{j-1}(\gamma^{-N}k)]U_{-}[S_{j-1}(\gamma^{-2N}k)]\mathcal{R}^{-1}[\Delta_{j-1}(\gamma^{-N}k)]\mathcal{R}^{-1}[\Delta_j(k)]\\
&=\mathcal{R}[\widetilde{\Delta}_j(k)]U_{-}[S_1(\gamma^{-Nj}k)]\mathcal{R}^{-1}[\widetilde{\Delta}_j(k)]\\
&=U_{-}[e^{-4i\widetilde{\Delta}_j(k)}S_1(\gamma^{-Nj}k)]. \tag*{(221)}
\end{aligned} U − [ S j + 1 ( k )] = R [ Δ j ( k )] R [ Δ j − 1 ( γ − N k )] U − [ S j − 1 ( γ − 2 N k )] R − 1 [ Δ j − 1 ( γ − N k )] R − 1 [ Δ j ( k )] = R [ Δ j ( k )] U − [ S 1 ( γ − N j k )] R − 1 [ Δ j ( k )] = U − [ e − 4 i Δ j ( k ) S 1 ( γ − N j k )] . (221)
with Δ ~ j ( k ) ≡ ∑ r = 1 j ( − 1 ) j − r Δ r ( k ) \widetilde{\Delta}_j(k)\equiv\sum_{r=1}^{j}(-1)^{j-r}\Delta_r(k) Δ j ( k ) ≡ ∑ r = 1 j ( − 1 ) j − r Δ r ( k ) an alternating sum. This leads to eq. (70) . For j = N + 1 j=N+1 j = N + 1 , as the common boundary of the ( N + 1 ) (N+1) ( N + 1 ) -th Stokes sector and the ( N + 2 ) (N+2) ( N + 2 ) -th Stokes sector is on the negative real axis, the rotation from U N + 1 ( k ) U_{N+1}(k) U N + 1 ( k ) to U N + 2 ( k ) U_{N+2}(k) U N + 2 ( k ) crosses no branch cuts and is given by eq. (69)
U + [ S N + 2 ( k ) ] = B U − [ S N + 1 ( γ − N k ) ] B , (222) U_{+}[S_{N+2}(k)]=BU_{-}[S_{N+1}(\gamma^{-N}k)]B, \tag*{(222)} U + [ S N + 2 ( k )] = B U − [ S N + 1 ( γ − N k )] B , (222)
which yields eq. (71) . On the lower half plane, the rotational symmetry representation (69) implies
U + [ S j + 1 ( k ) ] = R [ Δ j − 1 ( k ) ] U + [ S j ( γ − N k ) ] R − 1 [ Δ j − 1 ( k ) ] , N + 2 ≤ j ≤ 2 N + 1. (223) U_{+}[S_{j+1}(k)]=\mathcal{R}[\Delta_{j-1}(k)]U_{+}[S_j(\gamma^{-N}k)]\mathcal{R}^{-1}[\Delta_{j-1}(k)],\qquad N+2\leq j\leq2N+1. \tag*{(223)} U + [ S j + 1 ( k )] = R [ Δ j − 1 ( k )] U + [ S j ( γ − N k )] R − 1 [ Δ j − 1 ( k )] , N + 2 ≤ j ≤ 2 N + 1. (223)
Repeating the derivation for the upper half plane, we find
U + [ S N + 2 + l ( k ) ] = R [ − Δ ~ l ( k ) ] U + [ S N + 2 ( γ − N l k ) ] R − 1 [ − Δ ~ l ( k ) ] = U + [ e − 4 i Δ ~ l ( k ) S N + 2 ( γ − N l k ) ] . (224) \begin{aligned}
U_{+}[S_{N+2+l}(k)]&=\mathcal{R}[-\widetilde{\Delta}_l(k)]U_{+}[S_{N+2}(\gamma^{-Nl}k)]\mathcal{R}^{-1}[-\widetilde{\Delta}_l(k)]\\
&=U_{+}[e^{-4i\widetilde{\Delta}_l(k)}S_{N+2}(\gamma^{-Nl}k)]. \tag*{(224)}
\end{aligned} U + [ S N + 2 + l ( k )] = R [ − Δ l ( k )] U + [ S N + 2 ( γ − N l k )] R − 1 [ − Δ l ( k )] = U + [ e − 4 i Δ l ( k ) S N + 2 ( γ − N l k )] . (224)
With eqs. (70) and (71) the ( N + 2 ) (N+2) ( N + 2 ) -th Stokes constant can be written as
S N + 2 ( k ) = e 4 i Δ ~ N S 1 ( γ − N ( N + 1 ) k ) , (225) S_{N+2}(k)=e^{4i\widetilde{\Delta}_N}S_1(\gamma^{-N(N+1)}k), \tag*{(225)} S N + 2 ( k ) = e 4 i Δ N S 1 ( γ − N ( N + 1 ) k ) , (225)
which with eq. (C.7) proves the relation eq. (72) . Hence, through eqs. (70) –(72) all the Stokes constants are expressed in terms of the first Stokes constant with argument rotated and certain k k k -dependent phase factors.
Remnant symmetry
In this appendix, we will show explicitly that for a small but nonzero k k k , the first Stokes constant S 1 ( k ) S_1(k) S 1 ( k ) cannot be fixed merely by the rotational symmetry (70) –(72) , the time-reversal (85) and the single-valuedness of the exact mode χ k ( z ) \chi_k(z) χ k ( z ) . Specifically, we will prove this by constructing a nontrivial remnant transformation on the Stokes constants that is compatible with the symmetries and the single-valuedness.
We begin by considering a 2 π 2\pi 2 π propagation of the WKB amplitude vector V k ( z ) V_k(z) V k ( z ) along a closed contour around the origin, z → e 2 π i z z \to e^{2\pi i}z z → e 2 π i z . As the branch cuts have been specified in § 2.2 , both the exact mode χ k ( z ) \chi_k(z) χ k ( z ) and the WKB basis vector F k ( z ) F_k(z) F k ( z ) are single-valued
χ k ( e 2 π i z ) = χ k ( z ) , F k ( e 2 π i z ) = F k ( z ) . (226) \chi_k(e^{2\pi i}z)=\chi_k(z), \qquad F_k(e^{2\pi i}z)=F_k(z). \tag*{(226)} χ k ( e 2 π i z ) = χ k ( z ) , F k ( e 2 π i z ) = F k ( z ) . (226)
Given the decomposition χ k ( z ) = F k T ( z ) V k ( z ) \chi_k(z)=F_k^T(z)V_k(z) χ k ( z ) = F k T ( z ) V k ( z ) , we find that the WKB amplitude vector must also be single-valued
V k ( e 2 π i z ) = V k ( z ) . (227) V_k(e^{2\pi i}z)=V_k(z). \tag*{(227)} V k ( e 2 π i z ) = V k ( z ) . (227)
On the other hand, the WKB amplitude vector propagation along this contour is governed by a sequence of Stokes phenomena and branch cut factors, giving the single-valuedness constraint
U 2 N + 2 B 2 N U 2 N + 1 ⋯ B N + 1 U N + 2 U N + 1 B N ⋯ U 2 B 1 U 1 = I 2 × 2 . (228) U_{2N+2}\mathcal{B}_{2N}U_{2N+1}\cdots\mathcal{B}_{N+1}U_{N+2}U_{N+1}\mathcal{B}_{N}\cdots U_2\mathcal{B}_1U_1=\mathbb{I}_{2\times2}. \tag*{(228)} U 2 N + 2 B 2 N U 2 N + 1 ⋯ B N + 1 U N + 2 U N + 1 B N ⋯ U 2 B 1 U 1 = I 2 × 2 . (228)
We now show that the nontrivial remnant transformation on the Stokes constants
S j ( k ) → S ˉ j ( k ) = e 4 i f ( γ − N ( j − 1 ) k ) S j ( k ) , 1 ≤ j ≤ 2 N + 2 (229) S_j(k)\to\bar{S}_j(k)=e^{4if(\gamma^{-N(j-1)}k)}S_j(k),\qquad1\leq j\leq2N+2 \tag*{(229)} S j ( k ) → S ˉ j ( k ) = e 4 i f ( γ − N ( j − 1 ) k ) S j ( k ) , 1 ≤ j ≤ 2 N + 2 (229)
f ( k ) = i ∑ p ∈ o d d d p ( k N + 1 N ) p , d p ∈ R (230) f(k)=i\sum_{p\in\mathrm{odd}}d_p\left(k^{\frac{N+1}{N}}\right)^p,\qquad d_p\in\mathbb{R} \tag*{(230)} f ( k ) = i p ∈ odd ∑ d p ( k N N + 1 ) p , d p ∈ R (230)
with at least one nonzero d p d_p d p is compatible with the rotational symmetry, the time-reversal and the single-valuedness constraint (D.3).
The transformations (D.4) on S j ( k ) S_j(k) S j ( k ) are naturally induced from the shift S 1 ( k ) → S ˉ 1 ( k ) = e 4 i f ( k ) S 1 ( k ) S_1(k)\to\bar{S}_1(k)=e^{4if(k)}S_1(k) S 1 ( k ) → S ˉ 1 ( k ) = e 4 i f ( k ) S 1 ( k ) and the rotational symmetry representations (3.22-3.24). This derivation utilizes the alternating parity property f ( γ − N k ) = − f ( k ) f(\gamma^{-N}k)=-f(k) f ( γ − N k ) = − f ( k ) inherited from definition (D.5). Furthermore, because the phase factors e 4 i f ( γ − N ( j − 1 ) k ) e^{4if(\gamma^{-N(j-1)}k)} e 4 i f ( γ − N ( j − 1 ) k ) are real, the transformed Bogoliubov β k \beta_k β k remains purely imaginary, preserving time-reversal symmetry. Thus, the remnant transformation (D.4) is manifestly compatible with these symmetries.
To verify compatibility with the single-valuedness constraint, we apply the operator identity R [ x ] U ± [ S ] R − 1 [ x ] = U ± [ e ± 4 i x S ] \mathcal{R}[x]U_{\pm}[S]\mathcal{R}^{-1}[x]=U_{\pm}[e^{\pm4ix}S] R [ x ] U ± [ S ] R − 1 [ x ] = U ± [ e ± 4 i x S ] to express the transformed Stokes matrices in the conjugated form
U ˉ j = { R [ ( − 1 ) j f ( k ) ] U − [ S j ] R − 1 [ ( − 1 ) j f ( k ) ] , 1 ≤ j ≤ N + 1 , R [ ( − 1 ) j − 1 f ( k ) ] U + [ S j ] R − 1 [ ( − 1 ) j − 1 f ( k ) ] , N + 2 ≤ j ≤ 2 N + 2. (231) \bar{U}_j=
\begin{cases}
\mathcal{R}[(-1)^j f(k)]U_{-}[S_j]\mathcal{R}^{-1}[(-1)^j f(k)], & 1\leq j\leq N+1,\\
\mathcal{R}[(-1)^{j-1}f(k)]U_{+}[S_j]\mathcal{R}^{-1}[(-1)^{j-1}f(k)], & N+2\leq j\leq2N+2.
\end{cases} \tag*{(231)} U ˉ j = { R [( − 1 ) j f ( k )] U − [ S j ] R − 1 [( − 1 ) j f ( k )] , R [( − 1 ) j − 1 f ( k )] U + [ S j ] R − 1 [( − 1 ) j − 1 f ( k )] , 1 ≤ j ≤ N + 1 , N + 2 ≤ j ≤ 2 N + 2. (231)
Note for 1 ≤ j ≤ N 1\leq j\leq N 1 ≤ j ≤ N
U ˉ j + 1 B j U ˉ j = R [ ( − 1 ) j + 1 f ( k ) ] U − [ S j + 1 ] R − 1 [ ( − 1 ) j + 1 f ( k ) ] B j R [ ( − 1 ) j f ( k ) ] (232) \bar{U}_{j+1}\mathcal{B}_j\bar{U}_j=\mathcal{R}[(-1)^{j+1}f(k)]U_{-}[S_{j+1}]\mathcal{R}^{-1}[(-1)^{j+1}f(k)]\mathcal{B}_j\mathcal{R}[(-1)^j f(k)] \tag*{(232)} U ˉ j + 1 B j U ˉ j = R [( − 1 ) j + 1 f ( k )] U − [ S j + 1 ] R − 1 [( − 1 ) j + 1 f ( k )] B j R [( − 1 ) j f ( k )] (232)
× U − [ S j ] R − 1 [ ( − 1 ) j f ( k ) ] (233) \times U_{-}[S_j]\mathcal{R}^{-1}[(-1)^j f(k)] \tag*{(233)} × U − [ S j ] R − 1 [( − 1 ) j f ( k )] (233)
in which
R − 1 [ ( − 1 ) j + 1 f ( k ) ] B j R [ ( − 1 ) j f ( k ) ] = B j . (234) \mathcal{R}^{-1}[(-1)^{j+1}f(k)]\mathcal{B}_j\mathcal{R}[(-1)^j f(k)]=\mathcal{B}_j. \tag*{(234)} R − 1 [( − 1 ) j + 1 f ( k )] B j R [( − 1 ) j f ( k )] = B j . (234)
where we have used B [ Δ ] = − i R [ Δ ] B \mathcal{B}[\Delta] = -i\mathcal{R}[\Delta]B B [ Δ ] = − i R [ Δ ] B and B R [ x ] = R [ − x ] B B\mathcal{R}[x] = \mathcal{R}[-x]B B R [ x ] = R [ − x ] B . Hence, the propagation across the upper half plane is
U ˉ N + 1 B N ⋯ U ˉ 2 B 1 U ˉ 1 = R [ ( − 1 ) N + 1 f ( k ) ] U N + 1 B N ⋯ U 2 B 1 U 1 R − 1 [ − f ( k ) ] (235) \bar{U}_{N+1}\mathcal{B}_{N}\cdots\bar{U}_{2}\mathcal{B}_{1}\bar{U}_{1}=\mathcal{R}[(-1)^{N+1}f(k)]U_{N+1}\mathcal{B}_{N}\cdots U_{2}\mathcal{B}_{1}U_{1}\mathcal{R}^{-1}[-f(k)] \tag*{(235)} U ˉ N + 1 B N ⋯ U ˉ 2 B 1 U ˉ 1 = R [( − 1 ) N + 1 f ( k )] U N + 1 B N ⋯ U 2 B 1 U 1 R − 1 [ − f ( k )] (235)
and the propagation across the lower half plane is similarly given by
U ˉ 2 N + 2 B 2 N U ˉ 2 N + 1 ⋯ B N + 1 U ˉ N + 2 = R [ ( − 1 ) 2 N + 1 f ( k ) ] U 2 N + 2 B 2 N U 2 N + 1 × ⋯ × B N + 1 U N + 2 R − 1 [ ( − 1 ) N + 1 f ( k ) ] (236) \bar{U}_{2N+2}\mathcal{B}_{2N}\bar{U}_{2N+1}\cdots\mathcal{B}_{N+1}\bar{U}_{N+2}=\mathcal{R}[(-1)^{2N+1}f(k)]U_{2N+2}\mathcal{B}_{2N}U_{2N+1}\times\cdots
\times\mathcal{B}_{N+1}U_{N+2}\mathcal{R}^{-1}[(-1)^{N+1}f(k)] \tag*{(236)} U ˉ 2 N + 2 B 2 N U ˉ 2 N + 1 ⋯ B N + 1 U ˉ N + 2 = R [( − 1 ) 2 N + 1 f ( k )] U 2 N + 2 B 2 N U 2 N + 1 × ⋯ × B N + 1 U N + 2 R − 1 [( − 1 ) N + 1 f ( k )] (236)
which leads to
U ˉ 2 N + 2 B 2 N U ˉ 2 N + 1 ⋯ B N + 1 U ˉ N + 2 U ˉ N + 1 B N ⋯ U ˉ 2 B 1 U ˉ 1 = R [ ( − 1 ) 2 N + 1 f ( k ) ] U 2 N + 2 B 2 N U 2 N + 1 ⋯ B N + 1 U N + 2 U N + 1 B N ⋯ U 2 B 1 U 1 R − 1 [ − f ( k ) ] = R [ − f ( k ) ] I 2 × 2 R − 1 [ − f ( k ) ] = I 2 × 2 (237) \begin{aligned}
\bar{U}_{2N+2}\mathcal{B}_{2N}\bar{U}_{2N+1}\cdots\mathcal{B}_{N+1}\bar{U}_{N+2}\bar{U}_{N+1}\mathcal{B}_{N}\cdots\bar{U}_{2}\mathcal{B}_{1}\bar{U}_{1}
&=\mathcal{R}[(-1)^{2N+1}f(k)]U_{2N+2}\mathcal{B}_{2N}U_{2N+1}\cdots\mathcal{B}_{N+1}U_{N+2}U_{N+1}\mathcal{B}_{N}\cdots U_{2}\mathcal{B}_{1}U_{1}\mathcal{R}^{-1}[-f(k)]\\
&=\mathcal{R}[-f(k)]I_{2\times2}\mathcal{R}^{-1}[-f(k)]\\
&=I_{2\times2} \tag*{(237)}
\end{aligned} U ˉ 2 N + 2 B 2 N U ˉ 2 N + 1 ⋯ B N + 1 U ˉ N + 2 U ˉ N + 1 B N ⋯ U ˉ 2 B 1 U ˉ 1 = R [( − 1 ) 2 N + 1 f ( k )] U 2 N + 2 B 2 N U 2 N + 1 ⋯ B N + 1 U N + 2 U N + 1 B N ⋯ U 2 B 1 U 1 R − 1 [ − f ( k )] = R [ − f ( k )] I 2 × 2 R − 1 [ − f ( k )] = I 2 × 2 (237)
Hence, the transformed propagator of a 2 π 2\pi 2 π rotation reduces to identity, meaning that it is compatible with the single-valuedness constraint. Consequently, the combined constraints of rotational symmetry, time-reversal symmetry, and single-valuedness are insufficient to determine the first Stokes constant.
Non-analytic residue
In this appendix we derive the non-analytic residue expression (100) , which reads
θ ˉ ( k ) = Δ 1 ( k ) + Ψ 1 ( k ) = Δ 1 ( k ) e − i π n cos ( π n ) . (238) \bar{\theta}(k)=\Delta_{1}(k)+\Psi_{1}(k)=\frac{\Delta_{1}(k)e^{-\frac{i\pi}{n}}}{\cos\left(\frac{\pi}{n}\right)}. \tag*{(238)} θ ˉ ( k ) = Δ 1 ( k ) + Ψ 1 ( k ) = cos ( n π ) Δ 1 ( k ) e − n iπ . (238)
First, plugging ψ 1 = k 2 / n e i π / n \psi_{1}=k^{2/n}e^{i\pi/n} ψ 1 = k 2/ n e iπ / n , Δ 1 ( k ) = k ( n + 2 ) / n e i π / n d n \Delta_{1}(k)=k^{(n+2)/n}e^{i\pi/n}d_{n} Δ 1 ( k ) = k ( n + 2 ) / n e iπ / n d n (cf. eq. (B.2)) and the definition (B.7) for Ψ 1 ( k ) \Psi_{1}(k) Ψ 1 ( k ) into θ ˉ ( k ) \bar{\theta}(k) θ ˉ ( k ) yields
θ ˉ ( k ) = k n + 2 n e i π n d n + k n + 2 n e i π ( n + 2 ) 2 n ∑ m = 0 ∞ ( 1 / 2 m ) ( − 1 ) m m n − ( n 2 + 1 ) . (239) \bar{\theta}(k)=k^{\frac{n+2}{n}}e^{\frac{i\pi}{n}}d_{n}+k^{\frac{n+2}{n}}e^{\frac{i\pi(n+2)}{2n}}\sum_{m=0}^{\infty}\binom{1/2}{m}\frac{(-1)^{m}}{mn-\left(\frac{n}{2}+1\right)}. \tag*{(239)} θ ˉ ( k ) = k n n + 2 e n iπ d n + k n n + 2 e 2 n iπ ( n + 2 ) m = 0 ∑ ∞ ( m 1/2 ) mn − ( 2 n + 1 ) ( − 1 ) m . (239)
with
d n = Γ ( 3 2 ) Γ ( 1 + 1 n ) Γ ( 3 2 + 1 n ) . (240) d_{n}=\frac{\Gamma\left(\frac{3}{2}\right)\Gamma\left(1+\frac{1}{n}\right)}{\Gamma\left(\frac{3}{2}+\frac{1}{n}\right)}. \tag*{(240)} d n = Γ ( 2 3 + n 1 ) Γ ( 2 3 ) Γ ( 1 + n 1 ) . (240)
Let ν ≡ − ( 1 / 2 + 1 / n ) \nu\equiv-\left(1/2+1/n\right) ν ≡ − ( 1/2 + 1/ n ) which satisfies − 1 < ν < 0 -1<\nu<0 − 1 < ν < 0 for n ≥ 4 n\ge4 n ≥ 4 . The series in θ ˉ ( k ) \bar{\theta}(k) θ ˉ ( k ) can be written as
∑ m = 0 ∞ ( 1 / 2 m ) ( − 1 ) m m n − ( n 2 + 1 ) = 1 n ∑ m = 0 ∞ ( 1 / 2 m ) ( − 1 ) m m + ν (241) \sum_{m=0}^{\infty}\binom{1/2}{m}\frac{(-1)^{m}}{mn-\left(\frac{n}{2}+1\right)}=\frac{1}{n}\sum_{m=0}^{\infty}\binom{1/2}{m}\frac{(-1)^{m}}{m+\nu} \tag*{(241)} m = 0 ∑ ∞ ( m 1/2 ) mn − ( 2 n + 1 ) ( − 1 ) m = n 1 m = 0 ∑ ∞ ( m 1/2 ) m + ν ( − 1 ) m (241)
= 1 n Γ ( 3 2 ) Γ ( ν ) Γ ( ν + 3 2 ) (242) =\frac{1}{n}\frac{\Gamma\left(\frac{3}{2}\right)\Gamma(\nu)}{\Gamma\left(\nu+\frac{3}{2}\right)} \tag*{(242)} = n 1 Γ ( ν + 2 3 ) Γ ( 2 3 ) Γ ( ν ) (242)
As a result,
θ ˉ ( k ) = k n + 2 n e i π n d n [ 1 + e i π ( n + 2 2 n − 1 n ) Γ ( − 1 2 − 1 n ) Γ ( 3 2 + 1 n ) n Γ ( 1 − 1 n ) Γ ( 1 + 1 n ) ] . (243) \bar{\theta}(k)=k^{\frac{n+2}{n}}e^{\frac{i\pi}{n}}d_n\left[1+e^{i\pi\left(\frac{n+2}{2n}-\frac{1}{n}\right)}\frac{\Gamma\left(-\frac{1}{2}-\frac{1}{n}\right)\Gamma\left(\frac{3}{2}+\frac{1}{n}\right)}{n\Gamma\left(1-\frac{1}{n}\right)\Gamma\left(1+\frac{1}{n}\right)}\right]. \tag*{(243)} θ ˉ ( k ) = k n n + 2 e n iπ d n [ 1 + e iπ ( 2 n n + 2 − n 1 ) n Γ ( 1 − n 1 ) Γ ( 1 + n 1 ) Γ ( − 2 1 − n 1 ) Γ ( 2 3 + n 1 ) ] . (243)
Noting n Γ ( 1 + 1 / n ) = Γ ( 1 / n ) n\Gamma(1+1/n)=\Gamma(1/n) n Γ ( 1 + 1/ n ) = Γ ( 1/ n ) and Γ ( z ) Γ ( 1 − z ) = π / sin ( π z ) \Gamma(z)\Gamma(1-z)=\pi/\sin(\pi z) Γ ( z ) Γ ( 1 − z ) = π / sin ( π z ) , we find that eq. (E.6) transforms to
θ ˉ ( k ) = k n + 2 n d n e i π n ( 1 − i sin ( π / n ) cos ( π / n ) ) = e − i π n cos ( π n ) Δ 1 ( k ) . (244) \bar{\theta}(k)=k^{\frac{n+2}{n}}d_ne^{\frac{i\pi}{n}}\left(1-i\frac{\sin(\pi/n)}{\cos(\pi/n)}\right)=\frac{e^{-\frac{i\pi}{n}}}{\cos\left(\frac{\pi}{n}\right)}\Delta_1(k). \tag*{(244)} θ ˉ ( k ) = k n n + 2 d n e n iπ ( 1 − i cos ( π / n ) sin ( π / n ) ) = cos ( n π ) e − n iπ Δ 1 ( k ) . (244)
The generalized exponential integral
In this appendix, we derive the integral formulae eqs. (125) and (126) using the Stokes phenomena of the generalized exponential integral
E p ( z ) ≡ z p − 1 ∫ z ∞ d t e − t t p , z ≠ 0. (245) E_p(z)\equiv z^{p-1}\int_z^\infty dt\frac{e^{-t}}{t^p},\qquad z\ne0. \tag*{(245)} E p ( z ) ≡ z p − 1 ∫ z ∞ d t t p e − t , z = 0. (245)
The generalized exponential integral admits the following asymptotic expansions [36 ]
E p ( z ) ∼ e − z z ∑ s = 0 ∞ ( − 1 ) s ( p ) s z s , ∣ arg z ∣ < 3 π 2 , (246) E_p(z)\sim\frac{e^{-z}}{z}\sum_{s=0}^{\infty}\frac{(-1)^s(p)_s}{z^s},\qquad|\arg z|<\frac{3\pi}{2}, \tag*{(246)} E p ( z ) ∼ z e − z s = 0 ∑ ∞ z s ( − 1 ) s ( p ) s , ∣ arg z ∣ < 2 3 π , (246)
E p ( z ) ∼ e − z z ∑ s = 0 ∞ ( − 1 ) s ( p ) s z s + 2 π i e − i p π Γ ( p ) z p − 1 , π 2 < arg z < 7 π 2 , (247) E_p(z)\sim\frac{e^{-z}}{z}\sum_{s=0}^{\infty}\frac{(-1)^s(p)_s}{z^s}+2\pi i\frac{e^{-ip\pi}}{\Gamma(p)}z^{p-1},\qquad\frac{\pi}{2}<\arg z<\frac{7\pi}{2}, \tag*{(247)} E p ( z ) ∼ z e − z s = 0 ∑ ∞ z s ( − 1 ) s ( p ) s + 2 π i Γ ( p ) e − i p π z p − 1 , 2 π < arg z < 2 7 π , (247)
where ( p ) s ≡ Γ ( p + s ) / Γ ( p ) (p)_s\equiv\Gamma(p+s)/\Gamma(p) ( p ) s ≡ Γ ( p + s ) /Γ ( p ) . Interestingly, in the overlapping sector π / 2 < arg z < 3 π / 2 \pi/2<\arg z<3\pi/2 π /2 < arg z < 3 π /2 , the same analytic function E p ( z ) E_p(z) E p ( z ) has two different asymptotic expansions (F.2) and (F.3). This difference in the asymptotic expansions could be resolved using the Stokes phenomena, which says that when the argument of E p ( z ) E_p(z) E p ( z ) rotates counterclockwise and crosses the negative real axis from above, the integral acquires a Stokes term in its asymptotic expansion [37 ]
2 π i e − i p π Γ ( p ) z p − 1 . (248) 2\pi i\frac{e^{-ip\pi}}{\Gamma(p)}z^{p-1}. \tag*{(248)} 2 π i Γ ( p ) e − i p π z p − 1 . (248)
After turning on this Stokes term at the Stokes-line of e − z e^{-z} e − z at arg z = π \arg z=\pi arg z = π , this term remains exponentially suppressed compared to the ordinary asymptotic series (F.2) in the sector π < arg z < 3 π / 2 \pi<\arg z<3\pi/2 π < arg z < 3 π /2 , but becomes unsuppressed when z z z crosses the anti-Stokes line of e − z e^{-z} e − z at arg z = 3 π / 2 \arg z=3\pi/2 arg z = 3 π /2 . Therefore, in the overlapping sector π / 2 < arg z < 3 π / 2 \pi/2<\arg z<3\pi/2 π /2 < arg z < 3 π /2 , the two asymptotic expansions are not accurate enough to detect the exponentially suppressed Stokes term. This means only when one invokes a more accurate asymptotic expansion scheme, the exponentially improved asymptotic expansion, one can obtain the Stokes term in the expansion (see e.g. [37 ] ) as
E p ( z ) ∼ ˙ e − z z ∑ s = 0 ∞ ( − 1 ) s ( p ) s z s + Θ R ( arg z − π ) 2 π i e − i p π Γ ( p ) z p − 1 , − 3 π 2 < arg z < 7 π 2 , (249) E_p(z)\dot{\sim}\frac{e^{-z}}{z}\sum_{s=0}^{\infty}\frac{(-1)^s(p)_s}{z^s}+\Theta_{\mathbb{R}}(\arg z-\pi)2\pi i\frac{e^{-ip\pi}}{\Gamma(p)}z^{p-1},\qquad-\frac{3\pi}{2}<\arg z<\frac{7\pi}{2}, \tag*{(249)} E p ( z ) ∼ ˙ z e − z s = 0 ∑ ∞ z s ( − 1 ) s ( p ) s + Θ R ( arg z − π ) 2 π i Γ ( p ) e − i p π z p − 1 , − 2 3 π < arg z < 2 7 π , (249)
where Θ R ( x ) \Theta_{\mathrm{R}}(x) Θ R ( x ) denotes a step-function that jumps at x = 0 x=0 x = 0 with Θ R ( 0 ) = 1 / 2 \Theta_{\mathrm{R}}(0)=1/2 Θ R ( 0 ) = 1/2 and recall that ∼ ˙ \dot{\sim} ∼ ˙ denotes exponentially improved asymptotic expansion.
With the above results, we evaluate the exponentially improved asymptotic expansions of the integrals
I p ± ( z ) ≡ ∫ ∞ z d w e ± 2 i n / 2 + 1 w n / 2 + 1 w p (250) I_p^{\pm}(z) \equiv\int_{\infty}^{z} \mathrm{d}w \frac{e^{\pm\frac{2i}{n/2+1}w^{n/2+1}}}{w^p} \tag*{(250)} I p ± ( z ) ≡ ∫ ∞ z d w w p e ± n /2 + 1 2 i w n /2 + 1 (250)
via changes of variables (here the lower limit ∞ \infty ∞ is taken on the positive real axis).
For I p + ( z ) I_p^{+}(z) I p + ( z ) , we introduce the variable ζ ˉ ≡ − 2 i w N + 1 / ( N + 1 ) \bar{\zeta}\equiv-2iw^{N+1}/(N+1) ζ ˉ ≡ − 2 i w N + 1 / ( N + 1 ) . The upper integration limit maps to − ζ = − 2 i z N + 1 / ( N + 1 ) = − R ˉ e i arg ζ -\zeta=-2iz^{N+1}/(N+1)=-\bar{R}e^{i\arg\zeta} − ζ = − 2 i z N + 1 / ( N + 1 ) = − R ˉ e i a r g ζ , where R ˉ ≡ 2 R N + 1 / ( N + 1 ) \bar{R}\equiv2R^{N+1}/(N+1) R ˉ ≡ 2 R N + 1 / ( N + 1 ) with R = ∣ z ∣ R=|z| R = ∣ z ∣ is a large positive number and arg ζ ≤ π \arg\zeta\leq\pi arg ζ ≤ π . Consequently, we obtain
I p + ( z ) = 1 2 i ( 2 i N + 1 ) N + p N + 1 ( − 1 ) 1 − N + p N + 1 ∫ − ∞ − ζ d ζ ˉ e − ζ ˉ ζ ˉ N + p N + 1 = 1 2 i ( 2 i N + 1 ) N + p N + 1 ( − 1 ) 1 − N + p N + 1 [ − E N + p N + 1 ( − ζ ) ( − ζ ) N + p N + 1 − 1 ] ∼ ˙ e 2 i N + 1 z N + 1 ∑ s = 0 ∞ ( N + 1 ) s ( 2 i ) s + 1 ( N + p N + 1 s ) z ( N + 1 ) s + N + p . (251) \begin{aligned}
I_p^{+}(z)
&=\frac{1}{2i}\left(\frac{2i}{N+1}\right)^{\frac{N+p}{N+1}}(-1)^{1-\frac{N+p}{N+1}}\int_{-\infty}^{-\zeta}\mathrm{d}\bar{\zeta}\,\frac{e^{-\bar{\zeta}}}{\bar{\zeta}^{\frac{N+p}{N+1}}}\\
&=\frac{1}{2i}\left(\frac{2i}{N+1}\right)^{\frac{N+p}{N+1}}(-1)^{1-\frac{N+p}{N+1}}\left[-\frac{E_{\frac{N+p}{N+1}}(-\zeta)}{(-\zeta)^{\frac{N+p}{N+1}-1}}\right]\\
&\dot{\sim}e^{\frac{2i}{N+1}z^{N+1}}\sum_{s=0}^{\infty}\frac{(N+1)^s}{(2i)^{s+1}}\frac{\binom{\frac{N+p}{N+1}}{s}}{z^{(N+1)s+N+p}}. \tag*{(251)}
\end{aligned} I p + ( z ) = 2 i 1 ( N + 1 2 i ) N + 1 N + p ( − 1 ) 1 − N + 1 N + p ∫ − ∞ − ζ d ζ ˉ ζ ˉ N + 1 N + p e − ζ ˉ = 2 i 1 ( N + 1 2 i ) N + 1 N + p ( − 1 ) 1 − N + 1 N + p [ − ( − ζ ) N + 1 N + p − 1 E N + 1 N + p ( − ζ ) ] ∼ ˙ e N + 1 2 i z N + 1 s = 0 ∑ ∞ ( 2 i ) s + 1 ( N + 1 ) s z ( N + 1 ) s + N + p ( s N + 1 N + p ) . (251)
Applying the same change of variables to I p − ( z ) I_p^{-}(z) I p − ( z ) , we obtain
I p − ( z ) = 1 2 i ( 2 i N + 1 ) N + p N + 1 ∫ ∞ ζ d ζ ˉ e − ζ ˉ ζ N + p N + 1 = 1 2 i ( 2 i N + 1 ) N + p N + 1 [ − E N + p N + 1 ( ζ ) ζ N + p N + 1 − 1 ] . (252) \begin{aligned}
I_p^{-}(z)
&=\frac{1}{2i}\left(\frac{2i}{N+1}\right)^{\frac{N+p}{N+1}}\int_{\infty}^{\zeta}\mathrm{d}\bar{\zeta}\,\frac{e^{-\bar{\zeta}}}{\zeta^{\frac{N+p}{N+1}}}\\
&=\frac{1}{2i}\left(\frac{2i}{N+1}\right)^{\frac{N+p}{N+1}}\left[-\frac{E_{\frac{N+p}{N+1}}(\zeta)}{\zeta^{\frac{N+p}{N+1}-1}}\right]. \tag*{(252)}
\end{aligned} I p − ( z ) = 2 i 1 ( N + 1 2 i ) N + 1 N + p ∫ ∞ ζ d ζ ˉ ζ N + 1 N + p e − ζ ˉ = 2 i 1 ( N + 1 2 i ) N + 1 N + p [ − ζ N + 1 N + p − 1 E N + 1 N + p ( ζ ) ] . (252)
The term involving ζ \zeta ζ now acquires a Stokes contribution, as the integration path from ∞ \infty ∞ to ζ \zeta ζ crosses the negative real axis. Thus, the exponentially improved asymptotic expansion is
I p − ( z ) ∼ ˙ e − 2 i N + 1 z N + 1 ∑ s = 0 ∞ ( N + 1 ) s ( 2 i ) s + 1 ( − 1 ) s + 1 ( N + p N + 1 s ) z ( N + 1 ) s + N + p − Θ ( z ) π e − i π 2 N + p N + 1 Γ ( N + p N + 1 ) ( N + 1 2 ) N + p N + 1 . (253) I_p^{-}(z)\dot{\sim}e^{-\frac{2i}{N+1}z^{N+1}}\sum_{s=0}^{\infty}\frac{(N+1)^s}{(2i)^{s+1}}\frac{(-1)^{s+1}\binom{\frac{N+p}{N+1}}{s}}{z^{(N+1)s+N+p}}-\Theta(z)\frac{\pi e^{-\frac{i\pi}{2}\frac{N+p}{N+1}}}{\Gamma\left(\frac{N+p}{N+1}\right)\left(\frac{N+1}{2}\right)^{\frac{N+p}{N+1}}}. \tag*{(253)} I p − ( z ) ∼ ˙ e − N + 1 2 i z N + 1 s = 0 ∑ ∞ ( 2 i ) s + 1 ( N + 1 ) s z ( N + 1 ) s + N + p ( − 1 ) s + 1 ( s N + 1 N + p ) − Θ ( z ) Γ ( N + 1 N + p ) ( 2 N + 1 ) N + 1 N + p π e − 2 iπ N + 1 N + p . (253)
We summarize the asymptotic behaviors of these integrals as follows:
∫ ∞ z d w e ζ ( w ) w p ∼ ˙ e ζ ( z ) ∑ s = 0 ∞ E ˉ s , p z ( N + 1 ) s + N + p . (254) \int_{\infty}^{z}\mathrm{d}w\frac{e^{\zeta(w)}}{w^p}\dot{\sim}e^{\zeta(z)}\sum_{s=0}^{\infty}\frac{\bar{E}_{s,p}}{z^{(N+1)s+N+p}}. \tag*{(254)} ∫ ∞ z d w w p e ζ ( w ) ∼ ˙ e ζ ( z ) s = 0 ∑ ∞ z ( N + 1 ) s + N + p E ˉ s , p . (254)
∫ ∞ z d w e − ζ ( w ) w p ∼ ˙ e − ζ ( z ) ∑ s = 0 ∞ ( − 1 ) s + 1 E ˉ s , p z ( N + 1 ) s + N + p − Θ ( z ) S ˉ N + p N + 1 . (255) \int_{\infty}^{z}\mathrm{d}w\frac{e^{-\zeta(w)}}{w^p}\dot{\sim}e^{-\zeta(z)}\sum_{s=0}^{\infty}\frac{(-1)^{s+1}\bar{E}_{s,p}}{z^{(N+1)s+N+p}}-\Theta(z)\bar{S}_{\frac{N+p}{N+1}}. \tag*{(255)} ∫ ∞ z d w w p e − ζ ( w ) ∼ ˙ e − ζ ( z ) s = 0 ∑ ∞ z ( N + 1 ) s + N + p ( − 1 ) s + 1 E ˉ s , p − Θ ( z ) S ˉ N + 1 N + p . (255)
where the expansion coefficients and the Stokes constant are defined by
E ˉ s , p = ( N + 1 ) s ( 2 i ) s + 1 Γ ( N + p N + 1 + s ) Γ ( N + p N + 1 ) , S ˉ N + p N + 1 = π e − i π 2 N + p N + 1 Γ ( N + p N + 1 ) ( N + 1 2 ) N + p N + 1 . (256) \bar{E}_{s,p}=\frac{(N+1)^s}{(2i)^{s+1}}\frac{\Gamma\left(\frac{N+p}{N+1}+s\right)}{\Gamma\left(\frac{N+p}{N+1}\right)},\qquad
\bar{S}_{\frac{N+p}{N+1}}=\frac{\pi e^{-\frac{i\pi}{2}\frac{N+p}{N+1}}}{\Gamma\left(\frac{N+p}{N+1}\right)\left(\frac{N+1}{2}\right)^{\frac{N+p}{N+1}}}. \tag*{(256)} E ˉ s , p = ( 2 i ) s + 1 ( N + 1 ) s Γ ( N + 1 N + p ) Γ ( N + 1 N + p + s ) , S ˉ N + 1 N + p = Γ ( N + 1 N + p ) ( 2 N + 1 ) N + 1 N + p π e − 2 iπ N + 1 N + p . (256)
and recall that ζ ( z ) ≡ 2 i z N + 1 / ( N + 1 ) \zeta(z) \equiv2iz^{N+1}/(N+1) ζ ( z ) ≡ 2 i z N + 1 / ( N + 1 ) .
The asymmetry between the integrals I p + I_p^+ I p + and I p − I_p^- I p − arises from the distinct topologies of their integration paths in the complex ζ \zeta ζ -plane. As illustrated in Figure 6 , the mapping to the ζ \zeta ζ -variable reveals that the path for I p − I_p^- I p − intersects the Stokes line (the negative real axis), thereby enforcing the inclusion of the discontinuous Stokes term. In contrast, the path for I p + I_p^+ I p + does not encounter the Stokes line, thus avoids the Stokes term.
Figure 6. Integration paths in the complex w w w -plane and the mapped ζ ˉ \bar{\zeta} ζ ˉ -plane. The path for I p − I_p^- I p − (top right) crosses the Stokes line (dashed), resulting in the associated Stokes phenomena, whereas the path for I p + I_p^+ I p + (bottom right) does not cross the Stokes line.
Linear maps constructed from amplitude ODE
In this appendix, we derive explicit representations for the linear operators defined in eqs. (128) –(131) . While the abstract definitions suffice for formal developments of the solvable technique, solving the recursion relation to obtain the first Stokes constant requires the explicit action of these operators on the space of power series.
We begin with the linear maps induced by the generalized exponential integrals. Rewriting the constituent integrals acting on z − p z^{-p} z − p yields
∫ ∞ z d w e + ζ ( w ) w p ∼ e ζ ( z ) ∑ m = 0 ∞ E ˉ m − p − N N + 1 , p z m δ ( m − p − N N + 1 ) , (257) \int_{\infty}^{z} \mathrm{d}w\,\frac{e^{+\zeta(w)}}{w^{p}}\sim e^{\zeta(z)}\sum_{m=0}^{\infty}\frac{\bar{E}_{\frac{m-p-N}{N+1},p}}{z^{m}}\delta\left(\frac{m-p-N}{N+1}\right), \tag*{(257)} ∫ ∞ z d w w p e + ζ ( w ) ∼ e ζ ( z ) m = 0 ∑ ∞ z m E ˉ N + 1 m − p − N , p δ ( N + 1 m − p − N ) , (257)
∫ ∞ z d w e − ζ ( w ) w p ∼ e − ζ ( z ) ∑ m = 0 ∞ E ˉ m − p − N N + 1 , p z m ( − 1 ) m − p − N N + 1 + 1 δ ( m − p − N N + 1 ) − Θ ( z ) S ˉ N + p N + 1 , (258) \int_{\infty}^{z}\mathrm{d}w\,\frac{e^{-\zeta(w)}}{w^{p}}\sim e^{-\zeta(z)}\sum_{m=0}^{\infty}\frac{\bar{E}_{\frac{m-p-N}{N+1},p}}{z^{m}}(-1)^{\frac{m-p-N}{N+1}+1}\delta\left(\frac{m-p-N}{N+1}\right)-\Theta(z)\bar{S}_{\frac{N+p}{N+1}}, \tag*{(258)} ∫ ∞ z d w w p e − ζ ( w ) ∼ e − ζ ( z ) m = 0 ∑ ∞ z m E ˉ N + 1 m − p − N , p ( − 1 ) N + 1 m − p − N + 1 δ ( N + 1 m − p − N ) − Θ ( z ) S ˉ N + 1 N + p , (258)
where δ ( X ) \delta(X) δ ( X ) enforces the integer constraint arising from the dummy index shift
δ ( X ) = { 1 X ∈ Z ≥ 0 0 otherwise. (259) \delta(X)=\begin{cases}1 & X\in\mathbb{Z}_{\geq0}\\0 & \text{otherwise.}\end{cases} \tag*{(259)} δ ( X ) = { 1 0 X ∈ Z ≥ 0 otherwise. (259)
Consequently, the linear maps E ± \mathcal{E}_{\pm} E ± and the Stokes functional S \mathcal{S} S act on f ( z ) = ∑ m = 1 ∞ f m z − m ∈ P ( z ) f(z)=\sum_{m=1}^{\infty}f_mz^{-m}\in\mathcal{P}(z) f ( z ) = ∑ m = 1 ∞ f m z − m ∈ P ( z ) as
E + [ f ( z ) ] = ∑ m = 1 ∞ 1 z m + N ∑ p = 0 m − 1 f p + 1 E ˉ m − p − 1 N + 1 , p + 1 δ ( m − p − 1 N + 1 ) , (260) \mathcal{E}_{+}[f(z)]=\sum_{m=1}^{\infty}\frac{1}{z^{m+N}}\sum_{p=0}^{m-1}f_{p+1}\bar{E}_{\frac{m-p-1}{N+1},p+1}\delta\left(\frac{m-p-1}{N+1}\right), \tag*{(260)} E + [ f ( z )] = m = 1 ∑ ∞ z m + N 1 p = 0 ∑ m − 1 f p + 1 E ˉ N + 1 m − p − 1 , p + 1 δ ( N + 1 m − p − 1 ) , (260)
E − [ f ( z ) ] = ∑ m = 1 ∞ 1 z m + N ∑ p = 0 m − 1 f p + 1 ( − 1 ) m − p − 1 N + 1 + 1 E ˉ m − p − 1 N + 1 , p + 1 δ ( m − p − 1 N + 1 ) , (261) \mathcal{E}_{-}[f(z)]=\sum_{m=1}^{\infty}\frac{1}{z^{m+N}}\sum_{p=0}^{m-1}f_{p+1}(-1)^{\frac{m-p-1}{N+1}+1}\bar{E}_{\frac{m-p-1}{N+1},p+1}\delta\left(\frac{m-p-1}{N+1}\right), \tag*{(261)} E − [ f ( z )] = m = 1 ∑ ∞ z m + N 1 p = 0 ∑ m − 1 f p + 1 ( − 1 ) N + 1 m − p − 1 + 1 E ˉ N + 1 m − p − 1 , p + 1 δ ( N + 1 m − p − 1 ) , (261)
S [ f ( z ) ] = − ∑ p = 1 ∞ f p S ˉ N + p N + 1 . (262) \mathcal{S}[f(z)]=-\sum_{p=1}^{\infty}f_p\bar{S}_{\frac{N+p}{N+1}}. \tag*{(262)} S [ f ( z )] = − p = 1 ∑ ∞ f p S ˉ N + 1 N + p . (262)
Next, we consider multiplication by a fixed power series B ( z ) = ∑ m = 1 ∞ B m z − m ∈ P ( z ) B(z)=\sum_{m=1}^{\infty}B_mz^{-m}\in\mathcal{P}(z) B ( z ) = ∑ m = 1 ∞ B m z − m ∈ P ( z ) . The associated linear map B \mathcal{B} B acting on f ( z ) f(z) f ( z ) is given by
B [ f ( z ) ] = B ( z ) f ( z ) = ∑ p = 2 ∞ ( ∑ m + l = p B m f l z p ) . (263) \mathcal{B}[f(z)]=B(z)f(z)=\sum_{p=2}^{\infty}\left(\sum_{m+l=p}\frac{B_mf_l}{z^p}\right). \tag*{(263)} B [ f ( z )] = B ( z ) f ( z ) = p = 2 ∑ ∞ m + l = p ∑ z p B m f l . (263)
Finally, for the ordinary integral operator acting on g ( z ) = ∑ m = 2 ∞ g m z − m ∈ P ≥ 2 ( z ) g(z)=\sum_{m=2}^{\infty}g_mz^{-m}\in\mathcal{P}_{\geq2}(z) g ( z ) = ∑ m = 2 ∞ g m z − m ∈ P ≥ 2 ( z ) , we obtain
I [ g ( z ) ] ∼ ∑ p = 2 ∞ − g p p − 1 1 z p − 1 = − ∑ p = 1 ∞ g p + 1 p 1 z p ∈ P ( z ) . (264) \mathcal{I}[g(z)]\sim\sum_{p=2}^{\infty}\frac{-g_p}{p-1}\frac{1}{z^{p-1}}=-\sum_{p=1}^{\infty}\frac{g_{p+1}}{p}\frac{1}{z^p}\in\mathcal{P}(z). \tag*{(264)} I [ g ( z )] ∼ p = 2 ∑ ∞ p − 1 − g p z p − 1 1 = − p = 1 ∑ ∞ p g p + 1 z p 1 ∈ P ( z ) . (264)
Justifying V k \mathcal{V}_k V k
In this appendix we provide a rigorous justification for the form (115) of V k \mathcal{V}_k V k , the functional space of the exponentially improved asymptotic expansions of the WKB amplitudes.
In the following, we let z 0 = R z_0=R z 0 = R and z 1 = R exp ( π i / ( N + 1 ) ) z_1=R\exp(\pi i/(N+1)) z 1 = R exp ( π i / ( N + 1 )) with real R → ∞ R\to\infty R → ∞ be boundary points of the first Stokes sector. We start with the propagator version of the amplitude ODE, which is given by eqs. (34) –(36) as
V k ( z ) = U 0 ( k , z , z 0 ) V k ( z 0 ) , V k ( z 0 ) = ( 1 0 ) (265) \mathcal{V}_k(z)=U_0(k,z,z_0)\mathcal{V}_k(z_0),\qquad\mathcal{V}_k(z_0)=\begin{pmatrix}1\\0\end{pmatrix} \tag*{(265)} V k ( z ) = U 0 ( k , z , z 0 ) V k ( z 0 ) , V k ( z 0 ) = ( 1 0 ) (265)
U 0 ( k , z , z 0 ) = P [ I 2 × 2 exp ( ∫ ∞ z d w M 0 ( k , w ) ) ] . (266) U_0(k,z,z_0)=\mathbb{P}\left[\mathbb{I}_{2\times2}\exp\left(\int_{\infty}^{z}dw\,M_0(k,w)\right)\right]. \tag*{(266)} U 0 ( k , z , z 0 ) = P [ I 2 × 2 exp ( ∫ ∞ z d w M 0 ( k , w ) ) ] . (266)
where z z z denotes a point on the w w w -integration arc from z 0 z_0 z 0 to z 1 z_1 z 1 , as shown in Figure 7 .
Figure 7. The integration contour from z 0 z_0 z 0 to z z z in the complex w w w -plane used in the definition of the path-ordered exponential.
The propagator has a formal expansion
U 0 ( k , z , z 0 ) = P [ I 2 × 2 exp ( ∫ ∞ z d w M 0 ( k , w ) ) ] = I 2 × 2 + ∫ ∞ z d w 1 M 0 ( w 1 ) + ∫ ∞ z d w 1 M 0 ( w 1 ) ∫ ∞ w 1 d w 2 M 0 ( w 2 ) + ∫ ∞ z d w 1 M 0 ( w 1 ) ∫ ∞ w 1 d w 2 M 0 ( w 2 ) ∫ ∞ w 2 d w 3 M 0 ( w 3 ) + ⋯ (267) \begin{aligned}
U_0(k,z,z_0)&=\mathbb{P}\left[\mathbb{I}_{2\times2}\exp\left(\int_{\infty}^{z}dw\,M_0(k,w)\right)\right]\\
&=\mathbb{I}_{2\times2}+\int_{\infty}^{z}dw_1M_0(w_1)\\
&\quad+\int_{\infty}^{z}dw_1M_0(w_1)\int_{\infty}^{w_1}dw_2M_0(w_2)\\
&\quad+\int_{\infty}^{z}dw_1M_0(w_1)\int_{\infty}^{w_1}dw_2M_0(w_2)\int_{\infty}^{w_2}dw_3M_0(w_3)\\
&\quad+\cdots \tag*{(267)}
\end{aligned} U 0 ( k , z , z 0 ) = P [ I 2 × 2 exp ( ∫ ∞ z d w M 0 ( k , w ) ) ] = I 2 × 2 + ∫ ∞ z d w 1 M 0 ( w 1 ) + ∫ ∞ z d w 1 M 0 ( w 1 ) ∫ ∞ w 1 d w 2 M 0 ( w 2 ) + ∫ ∞ z d w 1 M 0 ( w 1 ) ∫ ∞ w 1 d w 2 M 0 ( w 2 ) ∫ ∞ w 2 d w 3 M 0 ( w 3 ) + ⋯ (267)
where due to the path-ordering in the integrations we require arg z 0 ≤ ⋯ ≤ arg w 3 ≤ arg w 2 ≤ arg w 1 ≤ arg z \arg z_0\le\cdots\le\arg w_3\le\arg w_2\le\arg w_1\le\arg z arg z 0 ≤ ⋯ ≤ arg w 3 ≤ arg w 2 ≤ arg w 1 ≤ arg z and we have suppressed the k k k dependence to avoid notational clutter. Using the definition (104) for B ± ( z ) B_{\pm}(z) B ± ( z ) we find
M 0 ( w ) = ( 0 e ζ ( w ) e 2 i θ ˉ ( k ) B − ( w ) e − ζ ( w ) e − 2 i θ ˉ ( k ) B + ( w ) 0 ) . (268) M_0(w)=
\begin{pmatrix}
0 & e^{\zeta(w)}e^{2i\bar{\theta}(k)}B_-(w)\\
e^{-\zeta(w)}e^{-2i\bar{\theta}(k)}B_+(w) & 0
\end{pmatrix}. \tag*{(268)} M 0 ( w ) = ( 0 e − ζ ( w ) e − 2 i θ ˉ ( k ) B + ( w ) e ζ ( w ) e 2 i θ ˉ ( k ) B − ( w ) 0 ) . (268)
Since B ± ( z ) ∈ P ( z ) B_{\pm}(z)\in\mathcal{P}(z) B ± ( z ) ∈ P ( z ) , the functional space of the exponentially improved asymptotic expansions of the first few terms in the formal expansion eq. (H.3) is determined by eqs. (111) and (112)
[ ∫ ∞ z d w 1 M 0 ( w 1 ) ] e . a . e . ∈ ( 0 e ζ ( z ) P ( z ) e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] 0 ) . (269) \left[\int_{\infty}^{z}dw_1M_0(w_1)\right]_{\mathrm{e.a.e.}}\in
\begin{pmatrix}
0 & e^{\zeta(z)}\mathcal{P}(z)\\
e^{-\zeta(z)}\mathcal{P}(z)+\Theta(z)\left[\mathcal{P}(z)+\mathbb{C}\right] & 0
\end{pmatrix}. \tag*{(269)} [ ∫ ∞ z d w 1 M 0 ( w 1 ) ] e.a.e. ∈ ( 0 e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] e ζ ( z ) P ( z ) 0 ) . (269)
[ ∫ ∞ z d w 1 M 0 ( w 1 ) ∫ ∞ w 1 d w 2 M 0 ( w 2 ) ] e . a . e . ∈ ( P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) 0 0 P ( z ) ) . (270) \left[\int_{\infty}^{z}dw_1M_0(w_1)\int_{\infty}^{w_1}dw_2M_0(w_2)\right]_{\mathrm{e.a.e.}}\in
\begin{pmatrix}
\mathcal{P}(z)+e^{\zeta(z)}\Theta(z)\mathcal{P}(z) & 0\\
0 & \mathcal{P}(z)
\end{pmatrix}. \tag*{(270)} [ ∫ ∞ z d w 1 M 0 ( w 1 ) ∫ ∞ w 1 d w 2 M 0 ( w 2 ) ] e.a.e. ∈ ( P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) 0 0 P ( z ) ) . (270)
[ ∫ ∞ z d w 1 M 0 ( w 1 ) ∫ ∞ w 1 d w 2 M 0 ( w 2 ) ∫ ∞ w 2 d w 3 M 0 ( w 3 ) ] e . a . e . ∈ ( 0 e ζ ( z ) P ( z ) e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] 0 ) (271) \left[\int_{\infty}^{z} \mathrm{d}w_{1}M_{0}(w_{1})\int_{\infty}^{w_{1}}\mathrm{d}w_{2}M_{0}(w_{2})\int_{\infty}^{w_{2}}\mathrm{d}w_{3}M_{0}(w_{3})\right]_{\mathrm{e.a.e.}}\in\left(\begin{array}{cc}0&e^{\zeta(z)\mathcal{P}(z)}\\ e^{-\zeta(z)\mathcal{P}(z)}+\Theta(z)\left[\mathcal{P}(z)+C\right]&0\end{array}\right) \tag*{(271)} [ ∫ ∞ z d w 1 M 0 ( w 1 ) ∫ ∞ w 1 d w 2 M 0 ( w 2 ) ∫ ∞ w 2 d w 3 M 0 ( w 3 ) ] e.a.e. ∈ ( 0 e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] e ζ ( z ) P ( z ) 0 ) (271)
Therefore, we conclude that the exponentially improved asymptotic expansions of all higher order terms are contained in the union of the functional spaces given by I 2 × 2 I_{2\times2} I 2 × 2 , that of the first order term (H.5), and that of the second order term (H.6). Hence, collecting the functional spaces of all order terms in the formal expansion (H.3), we obtain the functional space of the exponentially improved asymptotic expansion of the propagator:
[ U 0 ( z , z 0 ) ] e . a . e . ∈ ( 1 + P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) e ζ ( z ) P ( z ) e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] 1 + P ( z ) ) . (272) [U_{0}(z,z_{0})]_{\mathrm{e.a.e.}}\in\left(\begin{array}{cc}1+\mathcal{P}(z)+e^{\zeta(z)}\Theta(z)\mathcal{P}(z)&e^{\zeta(z)}\mathcal{P}(z)\\ e^{-\zeta(z)}\mathcal{P}(z)+\Theta(z)\left[\mathcal{P}(z)+C\right]&1+\mathcal{P}(z)\end{array}\right). \tag*{(272)} [ U 0 ( z , z 0 ) ] e.a.e. ∈ ( 1 + P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] e ζ ( z ) P ( z ) 1 + P ( z ) ) . (272)
The exponentially improved asymptotic expansion of the WKB amplitude propagated from V ( z 0 ) = ( 1 0 ) T V(z_{0})=\begin{pmatrix}1&0\end{pmatrix}^{T} V ( z 0 ) = ( 1 0 ) T is therefore
[ V k ( z ) ] e . a . e . ∈ ( 1 + P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] ) = V k (273) [V_{k}(z)]_{\mathrm{e.a.e.}}\in\left(\begin{array}{c}1+\mathcal{P}(z)+e^{\zeta(z)}\Theta(z)\mathcal{P}(z)\\ e^{-\zeta(z)}\mathcal{P}(z)+\Theta(z)\left[\mathcal{P}(z)+C\right]\end{array}\right)=V_{k} \tag*{(273)} [ V k ( z ) ] e.a.e. ∈ ( 1 + P ( z ) + e ζ ( z ) Θ ( z ) P ( z ) e − ζ ( z ) P ( z ) + Θ ( z ) [ P ( z ) + C ] ) = V k (273)
providing a rigorous justification for eq. (115) .
Non-analyticity cancellation details
In this appendix, we provide a detailed derivation from eq. (161) to eq. (162) , showing that for n ≥ 4 n\geq4 n ≥ 4 the non-analytic k k k factors in β k \beta_{k} β k all cancel out explicitly.
We start with eq. (161)
β k = ( − 1 ) N i N e 2 i Δ ~ N [ ∏ j = 1 N ( U + [ e 4 i ( − 1 ) j Δ ~ N S ~ 1 ( N + 2 + j ) ] B ) ] 21 . (274) \beta_{k}=(-1)^{N}i^{N}e^{2i\widetilde{\Delta}_{N}}\left[\prod_{j=1}^{N}\left(U_{+}\left[e^{4i(-1)^{j}\widetilde{\Delta}_{N}}\widetilde{S}_{1}^{(N+2+j)}\right]B\right)\right]_{21}. \tag*{(274)} β k = ( − 1 ) N i N e 2 i Δ N [ j = 1 ∏ N ( U + [ e 4 i ( − 1 ) j Δ N S 1 ( N + 2 + j ) ] B ) ] 21 . (274)
Using the first Stokes constant decomposition (119) we find
S ~ 1 ( N + 2 + j ) ( k ) = S 1 ( γ − N ( N + j + 1 ) k ) = e − 2 i ( − 1 ) N + j + 1 θ ˉ ( k ) T ~ 1 ( N + 2 + j ) ( k 2 ) , (275) \widetilde{S}_{1}^{(N+2+j)}(k)=S_{1}\left(\gamma^{-N(N+j+1)}k\right)=e^{-2i(-1)^{N+j+1}\bar{\theta}(k)}\widetilde{T}_{1}^{(N+2+j)}(k^{2}), \tag*{(275)} S 1 ( N + 2 + j ) ( k ) = S 1 ( γ − N ( N + j + 1 ) k ) = e − 2 i ( − 1 ) N + j + 1 θ ˉ ( k ) T 1 ( N + 2 + j ) ( k 2 ) , (275)
where we have used θ ˉ ( γ − N ( N + j + 1 ) k ) = ( − 1 ) N + j + 1 θ ˉ ( k ) \bar{\theta}\left(\gamma^{-N(N+j+1)}k\right)=(-1)^{N+j+1}\bar{\theta}(k) θ ˉ ( γ − N ( N + j + 1 ) k ) = ( − 1 ) N + j + 1 θ ˉ ( k ) . This leads to
β k = ( − 1 ) N i N e 2 i Δ ~ N [ ∏ j = 1 N ( U + [ e 4 i ( − 1 ) j x T ~ 1 ( N + 2 + j ) ] B ) ] 21 (276) \beta_{k}=(-1)^{N}i^{N}e^{2i\widetilde{\Delta}_{N}}\left[\prod_{j=1}^{N}\left(U_{+}\left[e^{4i(-1)^{j}x}\widetilde{T}_{1}^{(N+2+j)}\right]B\right)\right]_{21} \tag*{(276)} β k = ( − 1 ) N i N e 2 i Δ N [ j = 1 ∏ N ( U + [ e 4 i ( − 1 ) j x T 1 ( N + 2 + j ) ] B ) ] 21 (276)
x ≡ Δ ~ N ( k ) + ( − 1 ) N 2 θ ˉ ( k ) . (277) x\equiv\widetilde{\Delta}_{N}(k)+\frac{(-1)^{N}}{2}\bar{\theta}(k). \tag*{(277)} x ≡ Δ N ( k ) + 2 ( − 1 ) N θ ˉ ( k ) . (277)
We use matrix property eq. (C.2) and B R [ x ] B = R [ − x ] B\mathcal{R}[x]B=\mathcal{R}[-x] B R [ x ] B = R [ − x ] to get
U + [ e 4 i ( − 1 ) j x T ~ 1 ( N + 2 + j ) ] B = R [ ( − 1 ) j x ] U + [ T ~ 1 ( N + 2 + j ) ] B R − 1 [ ( − 1 ) j − 1 x ] , (278) U_{+}\left[e^{4i(-1)^{j}x}\widetilde{T}_{1}^{(N+2+j)}\right]B=\mathcal{R}\left[(-1)^{j}x\right]U_{+}\left[\widetilde{T}_{1}^{(N+2+j)}\right]B\mathcal{R}^{-1}\left[(-1)^{j-1}x\right], \tag*{(278)} U + [ e 4 i ( − 1 ) j x T 1 ( N + 2 + j ) ] B = R [ ( − 1 ) j x ] U + [ T 1 ( N + 2 + j ) ] B R − 1 [ ( − 1 ) j − 1 x ] , (278)
which leads to a cancellation in adjacent rotation matrices
∏ j = 1 N ( U + e 4 i ( − 1 ) j T ~ 1 ( N + 2 + j ) B ) = R [ ( − 1 ) N x ] ∏ j = 1 N ( U + [ T ~ 1 ( N + 2 + j ) ] B ) R − 1 [ x ] . \prod_{j=1}^{N}\left(U_{+}e^{4i(-1)^{j}\tilde{T}_{1}^{(N+2+j)}}B\right)=\mathcal{R}\left[(-1)^{N}x\right]\prod_{j=1}^{N}\left(U_{+}\left[\tilde{T}_{1}^{(N+2+j)}\right]B\right)\mathcal{R}^{-1}[x]. j = 1 ∏ N ( U + e 4 i ( − 1 ) j T ~ 1 ( N + 2 + j ) B ) = R [ ( − 1 ) N x ] j = 1 ∏ N ( U + [ T ~ 1 ( N + 2 + j ) ] B ) R − 1 [ x ] .
Therefore, eq. (I.3) becomes
β k = ( − 1 ) N i N e 2 i Δ ~ N [ R [ ( − 1 ) N x ] ∏ j = 1 N ( U + [ T ~ 1 ( N + 2 + j ) ] B ) R − 1 [ x ] ] 21 = ( − 1 ) N i N e 2 i [ Δ ~ N − ( 1 + ( − 1 ) N ) x ] [ ∏ j = 1 N ( U + [ T ~ 1 ( N + 2 + j ) ] B ) ] 21 , (279) \begin{aligned}
\beta_k&=(-1)^{N}i^{N}e^{2i\tilde{\Delta}_{N}}\left[\mathcal{R}\left[(-1)^{N}x\right]\prod_{j=1}^{N}\left(U_{+}\left[\tilde{T}_{1}^{(N+2+j)}\right]B\right)\mathcal{R}^{-1}[x]\right]_{21}\\
&=(-1)^{N}i^{N}e^{2i[\tilde{\Delta}_{N}-(1+(-1)^{N})x]}\left[\prod_{j=1}^{N}\left(U_{+}\left[\tilde{T}_{1}^{(N+2+j)}\right]B\right)\right]_{21}, \tag*{(279)}
\end{aligned} β k = ( − 1 ) N i N e 2 i Δ ~ N [ R [ ( − 1 ) N x ] j = 1 ∏ N ( U + [ T ~ 1 ( N + 2 + j ) ] B ) R − 1 [ x ] ] 21 = ( − 1 ) N i N e 2 i [ Δ ~ N − ( 1 + ( − 1 ) N ) x ] [ j = 1 ∏ N ( U + [ T ~ 1 ( N + 2 + j ) ] B ) ] 21 , (279)
which contains a non-analytic factor e 2 i [ Δ ~ N − ( 1 + ( − 1 ) N ) x ] e^{2i[\tilde{\Delta}_{N}-(1+(-1)^{N})x]} e 2 i [ Δ ~ N − ( 1 + ( − 1 ) N ) x ] and a finite product of power series analytic in k k k to leading order. Next, we show that the non-analytic phase factor is trivial. With the definitions (B.2) and (73) we obtain
Δ ~ N = k n + 2 n d n ∑ r = 1 N ( − 1 ) N − r e 2 r − 1 n i π = k n + 2 n d n 2 1 + ( − 1 ) N cos ( π n ) . (280) \tilde{\Delta}_{N}=k^{\frac{n+2}{n}}d_n\sum_{r=1}^{N}(-1)^{N-r}e^{\frac{2r-1}{n}i\pi}=\frac{k^{\frac{n+2}{n}}d_n}{2}\frac{1+(-1)^{N}}{\cos\left(\frac{\pi}{n}\right)}. \tag*{(280)} Δ ~ N = k n n + 2 d n r = 1 ∑ N ( − 1 ) N − r e n 2 r − 1 iπ = 2 k n n + 2 d n cos ( n π ) 1 + ( − 1 ) N . (280)
Also, using eq. (100) for the non-analytic residue we find
Δ ~ N − ( 1 + ( − 1 ) N ) x = − ( − 1 ) N Δ ~ N − 1 + ( − 1 ) N 2 θ ~ ( k ) = k n + 2 n d n 2 1 + ( − 1 ) N cos ( π n ) − 1 + ( − 1 ) N 2 k n + 2 n d n cos ( π n ) = 0 , (281) \begin{aligned}
\tilde{\Delta}_{N}-(1+(-1)^{N})x&=-(-1)^{N}\tilde{\Delta}_{N}-\frac{1+(-1)^{N}}{2}\tilde{\theta}(k)\\
&=\frac{k^{\frac{n+2}{n}}d_n}{2}\frac{1+(-1)^{N}}{\cos\left(\frac{\pi}{n}\right)}-\frac{1+(-1)^{N}}{2}\frac{k^{\frac{n+2}{n}}d_n}{\cos\left(\frac{\pi}{n}\right)}\\
&=0, \tag*{(281)}
\end{aligned} Δ ~ N − ( 1 + ( − 1 ) N ) x = − ( − 1 ) N Δ ~ N − 2 1 + ( − 1 ) N θ ~ ( k ) = 2 k n n + 2 d n cos ( n π ) 1 + ( − 1 ) N − 2 1 + ( − 1 ) N cos ( n π ) k n n + 2 d n = 0 , (281)
which shows the non-analyticity cancellation explicitly. Hence, eq. (I.6) becomes
β k = ( − 1 ) N i N [ ∏ j = 1 N ( U + [ T ~ 1 ( N + 2 + j ) ] B ) ] 21 , (282) \beta_k=(-1)^{N}i^{N}\left[\prod_{j=1}^{N}\left(U_{+}\left[\tilde{T}_{1}^{(N+2+j)}\right]B\right)\right]_{21}, \tag*{(282)} β k = ( − 1 ) N i N [ j = 1 ∏ N ( U + [ T ~ 1 ( N + 2 + j ) ] B ) ] 21 , (282)
which is eq. (162) we wanted to derive.
Closed form expression for c 1 c_1 c 1
The goal of this appendix is to derive the following closed form expression for the c 1 c_1 c 1 coefficient defined in the Bogoliubov β k \beta_k β k expansion (155) .24
c 1 ( n ) = i cos [ π 2 ( N + 1 ) ] Γ [ 1 2 ( N + 1 ) ] Γ [ 1 N + 1 ] Γ [ 3 2 ( N + 1 ) ] π ( N + 1 ) 2 ( 1 − 1 N + 1 ) Γ ( 1 N + 1 + 1 2 ) (283) c_1(n)=i\frac{\cos\left[\frac{\pi}{2(N+1)}\right]\Gamma\left[\frac{1}{2(N+1)}\right]\Gamma\left[\frac{1}{N+1}\right]\Gamma\left[\frac{3}{2(N+1)}\right]}{\sqrt{\pi}(N+1)^{2\left(1-\frac{1}{N+1}\right)}\Gamma\left(\frac{1}{N+1}+\frac{1}{2}\right)} \tag*{(283)} c 1 ( n ) = i π ( N + 1 ) 2 ( 1 − N + 1 1 ) Γ ( N + 1 1 + 2 1 ) cos [ 2 ( N + 1 ) π ] Γ [ 2 ( N + 1 ) 1 ] Γ [ N + 1 1 ] Γ [ 2 ( N + 1 ) 3 ] (283)
where N = n / 2 N=n/2 N = n /2 . The sketch of this derivation is as follows.
As a first step, in appendix J.1, we will split the explicit recursion relation (149) for G ( z ) G(z) G ( z ) into two sets of recursion relations, one for the topological limit terms G p ( 0 ) G_p^{(0)} G p ( 0 ) and the other for the soft correction terms G p ( 2 ) G_p^{(2)} G p ( 2 ) , where we recall that G p ( 0 ) G_p^{(0)} G p ( 0 ) and G p ( 2 ) G_p^{(2)} G p ( 2 ) are defined through the expansion G p = G p ( 0 ) + G p ( 2 ) k 2 + O ( k 4 ) G_p=G_p^{(0)}+G_p^{(2)}k^2+O(k^4) G p = G p ( 0 ) + G p ( 2 ) k 2 + O ( k 4 ) .
We then solve the explicit recursion relation for G p ( 0 ) G_p^{(0)} G p ( 0 ) in appendix J.2, which will yield the closed form expression (152) for T 1 ( 0 ) T_1^{(0)} T 1 ( 0 ) defined in eq. (141) . This will be used to reproduce the Stokes topological limit result β k = 0 = − i cot [ π / ( 2 ( N + 1 ) ) ] \beta_{k=0}=-i\cot[\pi/(2(N+1))] β k = 0 = − i cot [ π / ( 2 ( N + 1 ))] via the propagator product expression (162) .
In appendix J.3, we will solve the soft correction term G p ( 2 ) G_p^{(2)} G p ( 2 ) based on its recursion relation (J.11) and the solution (J.29) of G p ( 0 ) G_p^{(0)} G p ( 0 ) . Summing over G p ( 2 ) G_p^{(2)} G p ( 2 ) then yields the quadratic coefficient (coefficient of k 2 k^2 k 2 ) T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) in T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) as T 1 ( 2 ) = − ∑ p = 1 ∞ G p ( 2 ) T_1^{(2)}=-\sum_{p=1}^{\infty}G_p^{(2)} T 1 ( 2 ) = − ∑ p = 1 ∞ G p ( 2 ) .
Finally, in appendix J.4, we isolate the quadratic coefficient in the propagator product expression for β k \beta_k β k . This will express the quadratic coefficient of β k \beta_k β k , c 1 c_1 c 1 , in terms of T 1 ( 0 ) T_1^{(0)} T 1 ( 0 ) , T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) and trigonometric functions of N N N . Simplifying it leads to the closed form expression (J.1) for c 1 c_1 c 1 .
We repeat the explicit recursion relation (149) for G ( z ) G(z) G ( z ) and relevant definitions (150) and (F.12) here for reader’s convenience:
G p + 1 = { B p + 1 ( + ) S ˉ N + p + 1 N + 1 0 ≤ p ≤ N , B p + 1 ( + ) S ˉ N + p + 1 N + 1 − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s G s + 1 p ≥ N + 1 (284) G_{p+1}=
\begin{cases}
B_{p+1}^{(+)}\bar{S}_{\frac{N+p+1}{N+1}} & 0\le p\le N,\\
B_{p+1}^{(+)}\bar{S}_{\frac{N+p+1}{N+1}}-\sum_{m=N+1}^{p}\sum_{l=N+1}^{m}\sum_{s=0}^{l-N-1}K_{p;m,l,s}G_{s+1} & p\ge N+1
\end{cases} \tag*{(284)} G p + 1 = { B p + 1 ( + ) S ˉ N + 1 N + p + 1 B p + 1 ( + ) S ˉ N + 1 N + p + 1 − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s G s + 1 0 ≤ p ≤ N , p ≥ N + 1 (284)
K p ; m , l , s = B p − m + 1 ( + ) B m − l + 1 ( − ) m l S ˉ N + p + 1 N + 1 S ˉ N + l + 1 N + 1 δ ( l − s N + 1 − 1 ) (285) K_{p;m,l,s}=\frac{B_{p-m+1}^{(+)}B_{m-l+1}^{(-)}}{ml}\frac{\bar{S}_{\frac{N+p+1}{N+1}}}{\bar{S}_{\frac{N+l+1}{N+1}}}\delta\left(\frac{l-s}{N+1}-1\right) \tag*{(285)} K p ; m , l , s = m l B p − m + 1 ( + ) B m − l + 1 ( − ) S ˉ N + 1 N + l + 1 S ˉ N + 1 N + p + 1 δ ( N + 1 l − s − 1 ) (285)
S ˉ y = π e − i π 2 y Γ ( y ) ( N + 1 2 ) y , δ ( X ) = { 1 X ∈ Z ≥ 0 , 0 otherwise (286) \bar{S}_y=\frac{\pi e^{-\frac{i\pi}{2}y}}{\Gamma(y)\left(\frac{N+1}{2}\right)^y},\qquad
\delta(X)=
\begin{cases}
1 & X\in\mathbb{Z}_{\ge0},\\
0 & \text{otherwise}
\end{cases} \tag*{(286)} S ˉ y = Γ ( y ) ( 2 N + 1 ) y π e − 2 iπ y , δ ( X ) = { 1 0 X ∈ Z ≥ 0 , otherwise (286)
where the coefficients B m ( ± ) B_m^{(\pm)} B m ( ± ) are defined by eq. (104) . Throughout this appendix, we will be using the following index notations:
N + ≡ N + 1 , a ≡ N 2 ( N + 1 ) , d ≡ N − 1 N + 1 , φ ≡ π 2 ( N + 1 ) . (287) N_+\equiv N+1,\qquad a\equiv\frac{N}{2(N+1)},\qquad d\equiv\frac{N-1}{N+1},\qquad\varphi\equiv\frac{\pi}{2(N+1)}. \tag*{(287)} N + ≡ N + 1 , a ≡ 2 ( N + 1 ) N , d ≡ N + 1 N − 1 , φ ≡ 2 ( N + 1 ) π . (287)
Splitting the explicit recursion relation
To split the explicit recursion relation (J.2) in the soft momentum limit, we first expand the coefficients B m ( ± ) B_m^{(\pm)} B m ( ± ) to leading order of k 2 k^2 k 2 , which yields
B m ( ± ) = b m ( 0 ) + b m ( 2 ± ) k 2 + O ( k 4 ) (288) B_m^{(\pm)}=b_m^{(0)}+b_m^{(2\pm)}k^2+O(k^4) \tag*{(288)} B m ( ± ) = b m ( 0 ) + b m ( 2 ± ) k 2 + O ( k 4 ) (288)
b m ( 0 ) = N 2 δ m , 1 , b m ( 2 ± ) = ± i N 2 ( N − 1 ) δ m , N − N 2 δ m , 2 N + 1 (289) b_m^{(0)}=\frac{N}{2}\delta_{m,1},\qquad b_m^{(2\pm)}=\pm\frac{iN}{2(N-1)}\delta_{m,N}-\frac{N}{2}\delta_{m,2N+1} \tag*{(289)} b m ( 0 ) = 2 N δ m , 1 , b m ( 2 ± ) = ± 2 ( N − 1 ) i N δ m , N − 2 N δ m , 2 N + 1 (289)
where we have used eq. (104) and where δ m , p \delta_{m,p} δ m , p denotes the Kronecker delta. These expressions imply that the supports of b m ( 0 ) b_m^{(0)} b m ( 0 ) and b m ( 2 ± ) b_m^{(2\pm)} b m ( 2 ± ) are
supp [ b ( 0 ) ] = { 1 } , supp [ b ( 2 ± ) ] = { N , 2 N + 1 } , (290) \operatorname{supp}[b^{(0)}]=\{1\},\qquad\operatorname{supp}[b^{(2\pm)}]=\{N,2N+1\}, \tag*{(290)} supp [ b ( 0 ) ] = { 1 } , supp [ b ( 2 ± ) ] = { N , 2 N + 1 } , (290)
respectively.
Plugging the soft momentum expansions (J.6) for B m ( ± ) B_m^{(\pm)} B m ( ± ) and G p = G p ( 0 ) + G p ( 2 ) k 2 + O ( k 4 ) G_p=G_p^{(0)}+G_p^{(2)}k^2+O(k^4) G p = G p ( 0 ) + G p ( 2 ) k 2 + O ( k 4 ) into the explicit recursion relation (J.2) and collecting the constant terms gives the explicit recursion relation for the topological limit term G p ( 0 ) G_p^{(0)} G p ( 0 ) :
G p + 1 ( 0 ) = { b p + 1 ( 0 ) S ˉ N + p + 1 N + 1 , 0 ≤ p ≤ N , − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s ( 0 ) G s + 1 ( 0 ) , p ≥ N + 1 (291) G_{p+1}^{(0)}=
\begin{cases}
b_{p+1}^{(0)}\dfrac{\bar{S}_{N+p+1}}{N+1}, & 0\le p\le N,\\
-\sum_{m=N+1}^{p}\sum_{l=N+1}^{m}\sum_{s=0}^{l-N-1}K_{p;m,l,s}^{(0)}G_{s+1}^{(0)}, & p\ge N+1
\end{cases} \tag*{(291)} G p + 1 ( 0 ) = ⎩ ⎨ ⎧ b p + 1 ( 0 ) N + 1 S ˉ N + p + 1 , − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s ( 0 ) G s + 1 ( 0 ) , 0 ≤ p ≤ N , p ≥ N + 1 (291)
K p ; m , l , s ( 0 ) = b p − m + 1 ( 0 ) b m − l + 1 ( 0 ) m l S ˉ N + p + 1 / ( N + 1 ) S ˉ N + l + 1 / ( N + 1 ) δ ( l − s N + 1 − 1 ) , (292) K_{p;m,l,s}^{(0)}=\frac{b_{p-m+1}^{(0)}b_{m-l+1}^{(0)}}{ml}\frac{\bar{S}_{N+p+1}/(N+1)}{\bar{S}_{N+l+1}/(N+1)}\delta\left(\frac{l-s}{N+1}-1\right), \tag*{(292)} K p ; m , l , s ( 0 ) = m l b p − m + 1 ( 0 ) b m − l + 1 ( 0 ) S ˉ N + l + 1 / ( N + 1 ) S ˉ N + p + 1 / ( N + 1 ) δ ( N + 1 l − s − 1 ) , (292)
where in the second line of eq. (J.9) we have used the fact that b p + 1 ( 0 ) = 0 b_{p+1}^{(0)}=0 b p + 1 ( 0 ) = 0 for p ≥ N + 1 p\ge N+1 p ≥ N + 1 . Collecting the k 2 k^2 k 2 coefficients in eq. (J.2) instead yields the recursion relation for the soft correction term G p ( 2 ) G_p^{(2)} G p ( 2 ) :
G p + 1 ( 2 ) = { b p + 1 ( 2 + ) S ˉ N + p + 1 N + 1 , 0 ≤ p ≤ N , b p + 1 ( 2 + ) S ˉ N + p + 1 N + 1 − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 ( K p ; m , l , s ( 2 ) G s + 1 ( 0 ) + K p ; m , l , s ( 0 ) G s + 1 ( 2 ) ) , p ≥ N + 1 (293) G_{p+1}^{(2)}=
\begin{cases}
b_{p+1}^{(2+)}\dfrac{\bar{S}_{N+p+1}}{N+1}, & 0\le p\le N,\\
b_{p+1}^{(2+)}\dfrac{\bar{S}_{N+p+1}}{N+1}-\sum_{m=N+1}^{p}\sum_{l=N+1}^{m}\sum_{s=0}^{l-N-1}\left(K_{p;m,l,s}^{(2)}G_{s+1}^{(0)}+K_{p;m,l,s}^{(0)}G_{s+1}^{(2)}\right), & p\ge N+1
\end{cases} \tag*{(293)} G p + 1 ( 2 ) = ⎩ ⎨ ⎧ b p + 1 ( 2 + ) N + 1 S ˉ N + p + 1 , b p + 1 ( 2 + ) N + 1 S ˉ N + p + 1 − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 ( K p ; m , l , s ( 2 ) G s + 1 ( 0 ) + K p ; m , l , s ( 0 ) G s + 1 ( 2 ) ) , 0 ≤ p ≤ N , p ≥ N + 1 (293)
K p ; m , l , s ( 2 ) = b p − m + 1 ( 0 ) b m − l + 1 ( 2 − ) + b p − m + 1 ( 2 + ) b m − l + 1 ( 0 ) m l S ˉ N + p + 1 / ( N + 1 ) S ˉ N + l + 1 / ( N + 1 ) δ ( l − s N + 1 − 1 ) . (294) K_{p;m,l,s}^{(2)}=\frac{b_{p-m+1}^{(0)}b_{m-l+1}^{(2-)}+b_{p-m+1}^{(2+)}b_{m-l+1}^{(0)}}{ml}\frac{\bar{S}_{N+p+1}/(N+1)}{\bar{S}_{N+l+1}/(N+1)}\delta\left(\frac{l-s}{N+1}-1\right). \tag*{(294)} K p ; m , l , s ( 2 ) = m l b p − m + 1 ( 0 ) b m − l + 1 ( 2 − ) + b p − m + 1 ( 2 + ) b m − l + 1 ( 0 ) S ˉ N + l + 1 / ( N + 1 ) S ˉ N + p + 1 / ( N + 1 ) δ ( N + 1 l − s − 1 ) . (294)
As a result, eqs. (J.9) and (J.11) constitute the split explicit recursion relations. We will solve them in the next two appendices.
Reproducing the Stokes topological limit
In this appendix, we will reproduce the Stokes topological limit result
β k = 0 = − i cot [ π 2 ( N + 1 ) ] (295) \beta_{k=0}=-i\cot\left[\frac{\pi}{2(N+1)}\right] \tag*{(295)} β k = 0 = − i cot [ 2 ( N + 1 ) π ] (295)
by solving the explicit recursion relation (J.9) for G p ( 0 ) G_p^{(0)} G p ( 0 ) and using the propagator product expression (162) . In particular, we will first solve eq. (J.9) to obtain the finite sum ∑ m = 1 r G m ( 0 ) \sum_{m=1}^{r}G_m^{(0)} ∑ m = 1 r G m ( 0 ) , then take the r → ∞ r\to\infty r → ∞ limit to get the closed form expression for T 1 ( 0 ) T_1^{(0)} T 1 ( 0 ) defined in eq. (141) :
T 1 ( 0 ) = − lim r → ∞ ∑ m = 1 r G m ( 0 ) . (296) T_1^{(0)}=-\lim_{r\to\infty}\sum_{m=1}^{r}G_m^{(0)}. \tag*{(296)} T 1 ( 0 ) = − r → ∞ lim m = 1 ∑ r G m ( 0 ) . (296)
Plugging this into eq. (162) then yields eq. (J.13). A key step in this derivation is to show that the sequence G p ( 0 ) G_p^{(0)} G p ( 0 ) has a sparse support. This sparsity follows from the selection rules imposed by K p ; m , l , s ( 0 ) K_{p;m,l,s}^{(0)} K p ; m , l , s ( 0 ) in the recursion relation (J.9). Once this support has been determined explicitly, as in eq. (J.19), the original recursion reduces to a recursion for a nonzero subsequence and can be solved in closed form.
According to the first line of eq. (J.9), the only nonzero G p ( 0 ) G_p^{(0)} G p ( 0 ) element for 1 ≤ p ≤ N + 1 1\leq p\leq N+1 1 ≤ p ≤ N + 1 is
G 1 ( 0 ) = b 1 ( 0 ) S ˉ 1 = − i π N N + 1 = − 2 i π a , (297) G_1^{(0)}=b_1^{(0)}\bar{S}_1=-\frac{i\pi N}{N+1}=-2i\pi a, \tag*{(297)} G 1 ( 0 ) = b 1 ( 0 ) S ˉ 1 = − N + 1 iπ N = − 2 iπ a , (297)
where we have used eqs. (J.4), (J.5) and (J.7). For p ≥ N + 1 p\geq N+1 p ≥ N + 1 , the object K p ; m , l , s ( 0 ) K_{p;m,l,s}^{(0)} K p ; m , l , s ( 0 ) given by eq. (J.10) is nonzero if and only if b p − m + 1 ( 0 ) b_{p-m+1}^{(0)} b p − m + 1 ( 0 ) , b m − l + 1 ( 0 ) b_{m-l+1}^{(0)} b m − l + 1 ( 0 ) and δ ( ( l − s ) / ( N + 1 ) − 1 ) \delta((l-s)/(N+1)-1) δ (( l − s ) / ( N + 1 ) − 1 ) are all nonzero. Thus,
p − m + 1 ∈ supp [ b ( 0 ) ] , m − l + 1 ∈ supp [ b ( 0 ) ] , l − s N + 1 − 1 ∈ Z ≥ 0 , (298) p-m+1\in\operatorname{supp}[b^{(0)}],\qquad m-l+1\in\operatorname{supp}[b^{(0)}],\qquad\frac{l-s}{N+1}-1\in\mathbb{Z}_{\geq0}, \tag*{(298)} p − m + 1 ∈ supp [ b ( 0 ) ] , m − l + 1 ∈ supp [ b ( 0 ) ] , N + 1 l − s − 1 ∈ Z ≥ 0 , (298)
where the last condition follows from the definition of δ ( X ) \delta(X) δ ( X ) in eq. (J.4). Since supp [ b ( 0 ) ] = { 1 } \operatorname{supp}[b^{(0)}]=\{1\} supp [ b ( 0 ) ] = { 1 } , for a given p ≥ N + 1 p\geq N+1 p ≥ N + 1 , the first two conditions require m = l = p m=l=p m = l = p , and the third condition implies
p = s + ( j + 1 ) N + , j ∈ Z ≥ 0 (299) p=s+(j+1)N_+,\qquad j\in\mathbb{Z}_{\geq0} \tag*{(299)} p = s + ( j + 1 ) N + , j ∈ Z ≥ 0 (299)
with N + N_+ N + defined in eq. (J.5). The selection rule imposed by K p ; m , l , s ( 0 ) K_{p;m,l,s}^{(0)} K p ; m , l , s ( 0 ) can be summarized as
K p ; m , l , s ( 0 ) ≠ 0 iff m = l = p and p = s + ( j + 1 ) N + , j ∈ Z ≥ 0 . (300) K_{p;m,l,s}^{(0)}\ne0\quad\text{iff}\quad m=l=p\text{ and }p=s+(j+1)N_+,\ j\in\mathbb{Z}_{\geq0}. \tag*{(300)} K p ; m , l , s ( 0 ) = 0 iff m = l = p and p = s + ( j + 1 ) N + , j ∈ Z ≥ 0 . (300)
This selection rule implies that G p + 1 ( 0 ) G_{p+1}^{(0)} G p + 1 ( 0 ) can receive contributions only from G s + 1 ( 0 ) G_{s+1}^{(0)} G s + 1 ( 0 ) whose indices differ from p + 1 p+1 p + 1 by a positive integer multiple of N + N_+ N + . Furthermore, since G p ( 0 ) = ( − 2 π i a ) δ p , 1 G_p^{(0)}=(-2\pi ia)\delta_{p,1} G p ( 0 ) = ( − 2 π ia ) δ p , 1 for 1 ≤ p ≤ N + 1\leq p\leq N_+ 1 ≤ p ≤ N + , iterating the recursion relation (J.9) gives
supp [ G ( 0 ) ] ⊆ { 1 + r N + ∣ r ∈ Z ≥ 0 } . (301) \operatorname{supp}[G^{(0)}]\subseteq\{1+rN_+\mid r\in\mathbb{Z}_{\geq0}\}. \tag*{(301)} supp [ G ( 0 ) ] ⊆ { 1 + r N + ∣ r ∈ Z ≥ 0 } . (301)
Hence, the support of G p ( 0 ) G_p^{(0)} G p ( 0 ) is sparse, with successive nonzero coefficients separated by N + N_+ N + .
Given this sparse support, we may simplify the recursion relation (J.9) by considering the nonzero subsequence of G p ( 0 ) G_p^{(0)} G p ( 0 ) :
g r ( 0 ) ≡ G 1 + r N + ( 0 ) , r ≥ 0. (302) g_r^{(0)}\equiv G_{1+rN_+}^{(0)},\qquad r\geq0. \tag*{(302)} g r ( 0 ) ≡ G 1 + r N + ( 0 ) , r ≥ 0. (302)
We note
T 1 ( 0 ) = − ∑ m = 1 ∞ G m ( 0 ) = − ∑ m ∈ supp [ G ( 0 ) ] G m ( 0 ) = − ∑ r = 0 ∞ g r ( 0 ) . (303) T_1^{(0)}=-\sum_{m=1}^{\infty}G_m^{(0)}=-\sum_{m\in\operatorname{supp}[G^{(0)}]}G_m^{(0)}=-\sum_{r=0}^{\infty}g_r^{(0)}. \tag*{(303)} T 1 ( 0 ) = − m = 1 ∑ ∞ G m ( 0 ) = − m ∈ supp [ G ( 0 ) ] ∑ G m ( 0 ) = − r = 0 ∑ ∞ g r ( 0 ) . (303)
This motivates us to introduce the partial sum series
Y r ( 0 ) = ∑ t = 0 r g t ( 0 ) , r ≥ 0 (304) Y_r^{(0)} = \sum_{t=0}^{r} g_t^{(0)}, \qquad r \ge0 \tag*{(304)} Y r ( 0 ) = t = 0 ∑ r g t ( 0 ) , r ≥ 0 (304)
such that
T 1 ( 0 ) = − lim r → ∞ Y r ( 0 ) . (305) T_1^{(0)} = -\lim_{r\to\infty} Y_r^{(0)}. \tag*{(305)} T 1 ( 0 ) = − r → ∞ lim Y r ( 0 ) . (305)
Setting p = r N + p=rN_+ p = r N + and s = t N + s=tN_+ s = t N + in eq. (J.9) then leads to
g r ( 0 ) = { − 2 π i a r = 0 , − a 2 r 2 ∑ t = 0 r − 1 g t ( 0 ) r ≥ 1. (306) g_r^{(0)} =
\begin{cases}
-2\pi i a & r=0,\\
-\dfrac{a^2}{r^2}\sum_{t=0}^{r-1}g_t^{(0)} & r\ge1.
\end{cases} \tag*{(306)} g r ( 0 ) = ⎩ ⎨ ⎧ − 2 π ia − r 2 a 2 ∑ t = 0 r − 1 g t ( 0 ) r = 0 , r ≥ 1. (306)
where we have used the conditions (J.16) to reduce K p ; m , l , s ( 0 ) K_{p;m,l,s}^{(0)} K p ; m , l , s ( 0 ) . Since for r ≥ 1 r\ge1 r ≥ 1 , g r ( 0 ) = Y r ( 0 ) − Y r − 1 ( 0 ) g_r^{(0)}=Y_r^{(0)}-Y_{r-1}^{(0)} g r ( 0 ) = Y r ( 0 ) − Y r − 1 ( 0 ) , the recursion relation (J.24) for g r ( 0 ) g_r^{(0)} g r ( 0 ) translates to a recursion relation for Y r ( 0 ) Y_r^{(0)} Y r ( 0 ) :
Y r ( 0 ) = { − 2 π i a r = 0 , ( 1 − a 2 r 2 ) Y r − 1 ( 0 ) r ≥ 1. (307) Y_r^{(0)} =
\begin{cases}
-2\pi i a & r=0,\\
\left(1-\dfrac{a^2}{r^2}\right)Y_{r-1}^{(0)} & r\ge1.
\end{cases} \tag*{(307)} Y r ( 0 ) = ⎩ ⎨ ⎧ − 2 π ia ( 1 − r 2 a 2 ) Y r − 1 ( 0 ) r = 0 , r ≥ 1. (307)
Iterating this recursion relation yields
Y r ( 0 ) = [ ∏ j = 1 r ( 1 − a 2 j 2 ) ] Y 0 ( 0 ) = Γ ( r + 1 + a ) Γ ( r + 1 − a ) Γ ( 1 + a ) Γ ( 1 − a ) ( r ! ) 2 ( − 2 π i a ) , r ≥ 1 , (308) Y_r^{(0)}=\left[\prod_{j=1}^{r}\left(1-\dfrac{a^2}{j^2}\right)\right]Y_0^{(0)}
=\dfrac{\Gamma(r+1+a)\Gamma(r+1-a)}{\Gamma(1+a)\Gamma(1-a)(r!)^2}(-2\pi i a),\qquad r\ge1, \tag*{(308)} Y r ( 0 ) = [ j = 1 ∏ r ( 1 − j 2 a 2 ) ] Y 0 ( 0 ) = Γ ( 1 + a ) Γ ( 1 − a ) ( r ! ) 2 Γ ( r + 1 + a ) Γ ( r + 1 − a ) ( − 2 π ia ) , r ≥ 1 , (308)
where we have used the gamma function property ∏ j = 1 r ( j + x ) = Γ ( r + x + 1 ) / Γ ( x + 1 ) \prod_{j=1}^{r}(j+x)=\Gamma(r+x+1)/\Gamma(x+1) ∏ j = 1 r ( j + x ) = Γ ( r + x + 1 ) /Γ ( x + 1 ) to derive
∏ j = 1 r ( 1 − a 2 j 2 ) = ( ∏ j = 1 r j − a j ) ( ∏ j = 1 r j + a j ) = Γ ( r + 1 + a ) Γ ( r + 1 − a ) Γ ( 1 + a ) Γ ( 1 − a ) ( r ! ) 2 . (309) \prod_{j=1}^{r}\left(1-\dfrac{a^2}{j^2}\right)
=\left(\prod_{j=1}^{r}\dfrac{j-a}{j}\right)\left(\prod_{j=1}^{r}\dfrac{j+a}{j}\right)
=\dfrac{\Gamma(r+1+a)\Gamma(r+1-a)}{\Gamma(1+a)\Gamma(1-a)(r!)^2}. \tag*{(309)} j = 1 ∏ r ( 1 − j 2 a 2 ) = ( j = 1 ∏ r j j − a ) ( j = 1 ∏ r j j + a ) = Γ ( 1 + a ) Γ ( 1 − a ) ( r ! ) 2 Γ ( r + 1 + a ) Γ ( r + 1 − a ) . (309)
Given the second line of eq. (J.24), the definition (J.22) of Y r ( 0 ) Y_r^{(0)} Y r ( 0 ) , and the expression (J.26), we find
g r ( 0 ) = − a 2 r 2 Y r − 1 ( 0 ) = ( 2 π i a ) a 2 ( 1 + a ) r − 1 ( 1 − a ) r − 1 r 2 [ ( r − 1 ) ! ] 2 , r ≥ 1 , (310) g_r^{(0)}=-\dfrac{a^2}{r^2}Y_{r-1}^{(0)}=(2\pi i a)\dfrac{a^2(1+a)_{r-1}(1-a)_{r-1}}{r^2[(r-1)!]^2},\qquad r\ge1, \tag*{(310)} g r ( 0 ) = − r 2 a 2 Y r − 1 ( 0 ) = ( 2 π ia ) r 2 [( r − 1 )! ] 2 a 2 ( 1 + a ) r − 1 ( 1 − a ) r − 1 , r ≥ 1 , (310)
where we have used the Pochhammer symbol ( x ) s = Γ ( x + s ) / Γ ( x ) (x)_s=\Gamma(x+s)/\Gamma(x) ( x ) s = Γ ( x + s ) /Γ ( x ) . Since g r ( 0 ) g_r^{(0)} g r ( 0 ) is the nonzero subsequence of G p ( 0 ) G_p^{(0)} G p ( 0 ) , the solution to the explicit recursion relation eq. (J.9) is
G p ( 0 ) = { − 2 π i a p = 1 , 2 π i a a 2 ( 1 + a ) r − 1 ( 1 − a ) r − 1 r 2 [ ( r − 1 ) ! ] 2 p = 1 + r N + , r ≥ 1 , 0 otherwise . (311) G_p^{(0)}=
\begin{cases}
-2\pi i a & p=1,\\
2\pi i a\dfrac{a^2(1+a)_{r-1}(1-a)_{r-1}}{r^2[(r-1)!]^2} & p=1+rN_+,\quad r\ge1,\\
0 & \text{otherwise}.
\end{cases} \tag*{(311)} G p ( 0 ) = ⎩ ⎨ ⎧ − 2 π ia 2 π ia r 2 [( r − 1 )! ] 2 a 2 ( 1 + a ) r − 1 ( 1 − a ) r − 1 0 p = 1 , p = 1 + r N + , r ≥ 1 , otherwise . (311)
The topological limit term in T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) is then given by eq. (J.14) as
T 1 ( 0 ) = − lim r → ∞ Y r ( 0 ) = ( 2 π i a ) lim r → ∞ Γ ( r + 1 + a ) Γ ( r + 1 − a ) Γ ( 1 + a ) Γ ( 1 − a ) ( r ! ) 2 = 2 π i a Γ ( 1 + a ) Γ ( 1 − a ) . (312) T_1^{(0)}=-\lim_{r\to\infty}Y_r^{(0)}
=(2\pi i a)\lim_{r\to\infty}\dfrac{\Gamma(r+1+a)\Gamma(r+1-a)}{\Gamma(1+a)\Gamma(1-a)(r!)^2}
=\dfrac{2\pi i a}{\Gamma(1+a)\Gamma(1-a)}. \tag*{(312)} T 1 ( 0 ) = − r → ∞ lim Y r ( 0 ) = ( 2 π ia ) r → ∞ lim Γ ( 1 + a ) Γ ( 1 − a ) ( r ! ) 2 Γ ( r + 1 + a ) Γ ( r + 1 − a ) = Γ ( 1 + a ) Γ ( 1 − a ) 2 π ia . (312)
This can be further reduced using the gamma function identity Γ ( z ) Γ ( 1 − z ) = π / sin ( π z ) \Gamma(z)\Gamma(1-z)=\pi/\sin(\pi z) Γ ( z ) Γ ( 1 − z ) = π / sin ( π z ) :
T 1 ( 0 ) = 2 i sin ( π a ) = 2 i cos φ , (313) T_1^{(0)}=2i\sin(\pi a)=2i\cos\varphi, \tag*{(313)} T 1 ( 0 ) = 2 i sin ( π a ) = 2 i cos φ , (313)
where we have used eq. (J.5).
We now compute the Bogoliubov β k \beta_k β k using eq. (162) , which at the topological limit k = 0 k=0 k = 0 reads
β k = 0 = ( − 1 ) N i N [ X 0 N ] 21 , (314) \beta_{k=0}=(-1)^N i^N\left[X_0^N\right]_{21}, \tag*{(314)} β k = 0 = ( − 1 ) N i N [ X 0 N ] 21 , (314)
where
X 0 ≡ ( T 1 ( 0 ) 1 1 0 ) . (315) X_0\equiv
\begin{pmatrix}
T_1^{(0)} & 1\\
1 & 0
\end{pmatrix}. \tag*{(315)} X 0 ≡ ( T 1 ( 0 ) 1 1 0 ) . (315)
We compute X 0 N X_0^N X 0 N by solving a recursion relation for X 0 r ≥ 0 X_0^{r\geq0} X 0 r ≥ 0 . Since X 0 X_0 X 0 is symmetric, so is X 0 r X_0^r X 0 r . We therefore parameterize X 0 r X_0^r X 0 r as
X 0 r = ( a r b r b r c r ) , (316) X_0^r=
\begin{pmatrix}
a_r & b_r\\
b_r & c_r
\end{pmatrix}, \tag*{(316)} X 0 r = ( a r b r b r c r ) , (316)
then
X 0 r + 1 = ( a r b r b r c r ) ( T 1 ( 0 ) 1 1 0 ) = ( b r + T 1 ( 0 ) a r a r c r + T 1 ( 0 ) b r b r ) = ( a r + 1 b r + 1 b r + 1 c r + 1 ) . (317) X_0^{r+1}=
\begin{pmatrix}
a_r & b_r\\
b_r & c_r
\end{pmatrix}
\begin{pmatrix}
T_1^{(0)} & 1\\
1 & 0
\end{pmatrix}
=
\begin{pmatrix}
b_r+T_1^{(0)}a_r & a_r\\
c_r+T_1^{(0)}b_r & b_r
\end{pmatrix}
=
\begin{pmatrix}
a_{r+1} & b_{r+1}\\
b_{r+1} & c_{r+1}
\end{pmatrix}. \tag*{(317)} X 0 r + 1 = ( a r b r b r c r ) ( T 1 ( 0 ) 1 1 0 ) = ( b r + T 1 ( 0 ) a r c r + T 1 ( 0 ) b r a r b r ) = ( a r + 1 b r + 1 b r + 1 c r + 1 ) . (317)
This gives a set of recursion relations for the entries of X 0 r X_0^r X 0 r :
\begin{cases}
a_{r+1}=b_r+T_1^{(0)}a_r\\
b_{r+1}=a_r\\
c_{r+1}=b_r
\end{cases}
,\qquad r\geq0 \tag*{(318)}\end{cases}
with initial conditions given by ( X 0 ) 0 = I 2 × 2 (X_0)^0=\mathbb{I}_{2\times2} ( X 0 ) 0 = I 2 × 2 and eq. (J.33) as
\begin{cases}
a_0=1\\
b_0=0\\
c_0=1
\end{cases}
,\qquad
\begin{cases}
a_1=T_1^{(0)}\\
b_1=1\\
c_1=0
\end{cases}. \tag*{(319)}\end{cases}
From eq. (J.36) we obtain the recursion relation for b r b_r b r using eq. (J.31):
b r + 1 = b r − 1 + ( 2 i cos φ ) b r , r ≥ 1. (320) b_{r+1}=b_{r-1}+(2i\cos\varphi)b_r,\qquad r\geq1. \tag*{(320)} b r + 1 = b r − 1 + ( 2 i cos φ ) b r , r ≥ 1. (320)
This can be solved using the characteristic equation method. With the initial condition (J.37) we get
b r = λ + r − λ − r λ + − λ − , λ ± = i e ∓ i φ , r ≥ 0 , (321) b_r=\frac{\lambda_+^r-\lambda_-^r}{\lambda_+-\lambda_-},\qquad\lambda_\pm=ie^{\mp i\varphi},\qquad r\geq0, \tag*{(321)} b r = λ + − λ − λ + r − λ − r , λ ± = i e ∓ i φ , r ≥ 0 , (321)
where λ ± \lambda_\pm λ ± denote the characteristic roots of the recursion relation (J.38). This can be further simplified:
b r = i r − 1 e − i r φ − e i r φ e − i φ − e i φ = i r − 1 sin ( r φ ) sin φ , r ≥ 0. (322) b_r=i^{r-1}\frac{e^{-ir\varphi}-e^{ir\varphi}}{e^{-i\varphi}-e^{i\varphi}}=i^{r-1}\frac{\sin(r\varphi)}{\sin\varphi},\qquad r\geq0. \tag*{(322)} b r = i r − 1 e − i φ − e i φ e − i r φ − e i r φ = i r − 1 sin φ sin ( r φ ) , r ≥ 0. (322)
As a result, the Bogoliubov β k \beta_k β k at the topological limit is obtained from eq. (J.32) as
β k = 0 = ( − 1 ) N i N b N = − i sin ( N φ ) sin φ . (323) \beta_{k=0}=(-1)^N i^N b_N=-i\frac{\sin(N\varphi)}{\sin\varphi}. \tag*{(323)} β k = 0 = ( − 1 ) N i N b N = − i sin φ sin ( N φ ) . (323)
Restoring φ = π / ( 2 ( N + 1 ) ) \varphi=\pi/(2(N+1)) φ = π / ( 2 ( N + 1 )) as defined in eq. (J.5), we express β k = 0 \beta_{k=0} β k = 0 in terms of N N N as
β k = 0 = − i cot [ π 2 ( N + 1 ) ] . (324) \beta_{k=0}=-i\cot\left[\frac{\pi}{2(N+1)}\right]. \tag*{(324)} β k = 0 = − i cot [ 2 ( N + 1 ) π ] . (324)
This reproduces the Stokes topological limit result in [9 ] .
Quadratic coefficient of T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 )
This appendix is devoted to deriving the closed form expression (153) for T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) , the quadratic coefficient in the soft momentum expansion (141) for T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) . Analogous to the logic of appendix J.2, we will first solve the explicit recursion relation (J.11) for G p ( 2 ) G_p^{(2)} G p ( 2 ) , then use the propagator product expression (162) . Specifically, we will first solve eq. (J.11) to get the finite sum ∑ t = 1 r G t ( 2 ) \sum_{t=1}^{r}G_t^{(2)} ∑ t = 1 r G t ( 2 ) then take the r → ∞ r\to\infty r → ∞ limit to obtain the closed form expression for T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) given by
T 1 ( 2 ) = − lim r → ∞ ∑ t = 1 r G t ( 2 ) . (325) T_1^{(2)}=-\lim_{r\to\infty}\sum_{t=1}^{r}G_t^{(2)}. \tag*{(325)} T 1 ( 2 ) = − r → ∞ lim t = 1 ∑ r G t ( 2 ) . (325)
The key to solving G p ( 2 ) G_p^{(2)} G p ( 2 ) is also similar to that in appendix J.2. We will show that the sequence G p ( 2 ) G_p^{(2)} G p ( 2 ) has a sparse support where the sparsity follows from the selection rules imposed by K p ; m , l , s ( 2 ) K_{p;m,l,s}^{(2)} K p ; m , l , s ( 2 ) . By determining this support we will reduce the original recursion relation (J.11) to that of a nonzero subsequence, which can be solved in closed form. The main difference between the derivation here and that in the previous appendix arises when taking r → ∞ r\to\infty r → ∞ in eq. (J.43). Because the recursion relation for the nonzero subsequence of G p ( 2 ) G_p^{(2)} G p ( 2 ) is more involved than that of G p ( 0 ) G_p^{(0)} G p ( 0 ) , we will need to invoke hypergeometric function properties to obtain the closed form as r → ∞ r\to\infty r → ∞ in eq. (J.43).
We start with working out the support of G p ( 2 ) G_p^{(2)} G p ( 2 ) by reducing its recursion relation (J.11). For 1 ≤ p + 1 ≤ N + 1 1\leq p+1\leq N+1 1 ≤ p + 1 ≤ N + 1 , eq. (J.11) gives
G p + 1 ( 2 ) = b p + 1 ( 2 + ) S ˉ N + p + 1 N + p + 1 / ( N + 1 ) = { 0 1 ≤ p + 1 ≤ N − 1 , b N ( 2 + ) S ˉ 2 N 2 N / ( N + 1 ) p + 1 = N , 0 p + 1 = N + 1 (326) G_{p+1}^{(2)}=b_{p+1}^{(2+)}\bar{S}_{N+p+1}^{\phantom{N+p+1}/(N+1)}=
\begin{cases}
0 & 1\leq p+1\leq N-1,\\
b_N^{(2+)}\bar{S}_{2N}^{\phantom{2N}/(N+1)} & p+1=N,\\
0 & p+1=N+1
\end{cases} \tag*{(326)} G p + 1 ( 2 ) = b p + 1 ( 2 + ) S ˉ N + p + 1 N + p + 1 / ( N + 1 ) = ⎩ ⎨ ⎧ 0 b N ( 2 + ) S ˉ 2 N 2 N / ( N + 1 ) 0 1 ≤ p + 1 ≤ N − 1 , p + 1 = N , p + 1 = N + 1 (326)
where we have used eq. (J.7). For p + 1 ≥ N + 2 p+1\geq N+2 p + 1 ≥ N + 2 , the second line of eq. (J.11) implies that G p + 1 ( 2 ) G_{p+1}^{(2)} G p + 1 ( 2 ) is equal to a source term minus a propagator term:
G p + 1 ( 2 ) = b p + 1 ( 2 + ) S ˉ N + p + 1 N + p + 1 / ( N + 1 ) − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s ( 2 ) G s + 1 ( 0 ) ⏟ ≡ s o u r c e t e r m − ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s ( 0 ) G s + 1 ( 2 ) ⏟ ≡ p r o p a g a t o r t e r m . (327) G_{p+1}^{(2)}=b_{p+1}^{(2+)}\bar{S}_{N+p+1}^{\phantom{N+p+1}/(N+1)}
-\underbrace{\sum_{m=N+1}^{p}\sum_{l=N+1}^{m}\sum_{s=0}^{l-N-1}K_{p;m,l,s}^{(2)}G_{s+1}^{(0)}}_{\equiv\ \mathrm{source\ term}}
-\underbrace{\sum_{m=N+1}^{p}\sum_{l=N+1}^{m}\sum_{s=0}^{l-N-1}K_{p;m,l,s}^{(0)}G_{s+1}^{(2)}}_{\equiv\ \mathrm{propagator\ term}}. \tag*{(327)} G p + 1 ( 2 ) = b p + 1 ( 2 + ) S ˉ N + p + 1 N + p + 1 / ( N + 1 ) − ≡ source term m = N + 1 ∑ p l = N + 1 ∑ m s = 0 ∑ l − N − 1 K p ; m , l , s ( 2 ) G s + 1 ( 0 ) − ≡ propagator term m = N + 1 ∑ p l = N + 1 ∑ m s = 0 ∑ l − N − 1 K p ; m , l , s ( 0 ) G s + 1 ( 2 ) . (327)
In this expression, the support of the source term can be determined without knowledge of G s + 1 ( 2 ) G_{s+1}^{(2)} G s + 1 ( 2 ) , whereas the support of the propagator term must be determined using the support of the source term and the selection rule (J.18) imposed by K p ; m , l , s ( 0 ) K_{p;m,l,s}^{(0)} K p ; m , l , s ( 0 ) .
We first consider the source term in the range p ≥ N + 1 p \ge N+1 p ≥ N + 1 . According to eq. (J.19), G s + 1 ( 0 ) ≠ 0 G_{s+1}^{(0)} \ne0 G s + 1 ( 0 ) = 0 requires
s = t N + , t ∈ Z ≥ 0 , (328) s=tN_{+},\qquad t\in\mathbb{Z}_{\ge0}, \tag*{(328)} s = t N + , t ∈ Z ≥ 0 , (328)
where N + = N + 1 N_{+}=N+1 N + = N + 1 was defined in eq. (J.5). Furthermore, given eq. (J.12), K p ; m , l , s ( 2 ) K_{p;m,l,s}^{(2)} K p ; m , l , s ( 2 ) splits into two terms
K p ; m , l , s ( 2 ) = K p ; m , l , s ( 2 + ) + K p ; m , l , s ( 2 − ) (329) K_{p;m,l,s}^{(2)}=K_{p;m,l,s}^{(2+)}+K_{p;m,l,s}^{(2-)} \tag*{(329)} K p ; m , l , s ( 2 ) = K p ; m , l , s ( 2 + ) + K p ; m , l , s ( 2 − ) (329)
K p ; m , l , s ( 2 + ) ≡ b p − m + 1 ( 2 + ) b m − l + 1 ( 0 ) m l S ˉ N + p + 1 N + 1 S ˉ N + l + 1 N + 1 δ ( l − s N + 1 − 1 ) . (330) K_{p;m,l,s}^{(2+)}\equiv\frac{b_{p-m+1}^{(2+)}b_{m-l+1}^{(0)}}{ml}\frac{\bar{S}_{N+p+1}^{N+1}}{\bar{S}_{N+l+1}^{N+1}}\delta\left(\frac{l-s}{N+1}-1\right). \tag*{(330)} K p ; m , l , s ( 2 + ) ≡ m l b p − m + 1 ( 2 + ) b m − l + 1 ( 0 ) S ˉ N + l + 1 N + 1 S ˉ N + p + 1 N + 1 δ ( N + 1 l − s − 1 ) . (330)
K p ; m , l , s ( 2 − ) ≡ b p − m + 1 ( 0 ) b m − l + 1 ( 2 − ) m l S ˉ N + p + 1 N + 1 S ˉ N + l + 1 N + 1 δ ( l − s N + 1 − 1 ) . (331) K_{p;m,l,s}^{(2-)}\equiv\frac{b_{p-m+1}^{(0)}b_{m-l+1}^{(2-)}}{ml}\frac{\bar{S}_{N+p+1}^{N+1}}{\bar{S}_{N+l+1}^{N+1}}\delta\left(\frac{l-s}{N+1}-1\right). \tag*{(331)} K p ; m , l , s ( 2 − ) ≡ m l b p − m + 1 ( 0 ) b m − l + 1 ( 2 − ) S ˉ N + l + 1 N + 1 S ˉ N + p + 1 N + 1 δ ( N + 1 l − s − 1 ) . (331)
Similar to eq. (J.17), the common delta factor δ ( ( l − s ) / N + − 1 ) \delta((l-s)/N_{+}-1) δ (( l − s ) / N + − 1 ) is nonzero when
l = s + ( j + 1 ) N + = ( t + j + 1 ) N + , j ∈ Z ≥ 0 . (332) l=s+(j+1)N_{+}=(t+j+1)N_{+},\qquad j\in\mathbb{Z}_{\ge0}. \tag*{(332)} l = s + ( j + 1 ) N + = ( t + j + 1 ) N + , j ∈ Z ≥ 0 . (332)
Moreover, using the supports of b ( 0 ) b^{(0)} b ( 0 ) and b ( 2 ± ) b^{(2\pm)} b ( 2 ± ) given in eq. (J.7) we find that
K p ; m , l , s ( 2 + ) ≠ 0 i f f m = l , p − l + 1 ∈ { N , 2 N + 1 } , (333) K_{p;m,l,s}^{(2+)}\ne0\quad\mathrm{iff}\quad m=l,\quad p-l+1\in\{N,\,2N+1\}, \tag*{(333)} K p ; m , l , s ( 2 + ) = 0 iff m = l , p − l + 1 ∈ { N , 2 N + 1 } , (333)
K p ; m , l , s ( 2 − ) ≠ 0 i f f m = p , m − l + 1 ∈ { N , 2 N + 1 } . (334) K_{p;m,l,s}^{(2-)}\ne0\quad\mathrm{iff}\quad m=p,\quad m-l+1\in\{N,\,2N+1\}. \tag*{(334)} K p ; m , l , s ( 2 − ) = 0 iff m = p , m − l + 1 ∈ { N , 2 N + 1 } . (334)
Both cases require
p + 1 ∈ { l + N , l + 2 N + 1 } . (335) p+1\in\{l+N,\,l+2N+1\}. \tag*{(335)} p + 1 ∈ { l + N , l + 2 N + 1 } . (335)
Using l = ( t + j + 1 ) N + l=(t+j+1)N_{+} l = ( t + j + 1 ) N + obtained in eq. (J.50) and noting 2 N + 1 = N + N + 2N+1=N+N_{+} 2 N + 1 = N + N + , we obtain
p + 1 ∈ { N + ( t + j + 1 ) N + , N + ( t + j + 2 ) N + } . (336) p+1\in\{N+(t+j+1)N_{+},\,N+(t+j+2)N_{+}\}. \tag*{(336)} p + 1 ∈ { N + ( t + j + 1 ) N + , N + ( t + j + 2 ) N + } . (336)
Consequently, the K ( 2 ) G ( 0 ) K^{(2)}G^{(0)} K ( 2 ) G ( 0 ) contribution has support
supp [ ∑ m = N + 1 p ∑ l = N + 1 m ∑ s = 0 l − N − 1 K p ; m , l , s ( 2 ) G s + 1 ( 0 ) ] ⊆ { N + r N + ∣ r ∈ Z ≥ 1 } . (337) \operatorname{supp}\left[\sum_{m=N+1}^{p}\sum_{l=N+1}^{m}\sum_{s=0}^{l-N-1}K_{p;m,l,s}^{(2)}G_{s+1}^{(0)}\right]\subseteq\{N+rN_{+}\mid r\in\mathbb{Z}_{\ge1}\}. \tag*{(337)} supp [ m = N + 1 ∑ p l = N + 1 ∑ m s = 0 ∑ l − N − 1 K p ; m , l , s ( 2 ) G s + 1 ( 0 ) ] ⊆ { N + r N + ∣ r ∈ Z ≥ 1 } . (337)
where r ≥ 1 r\ge1 r ≥ 1 because eq. (J.45) is only valid for p + 1 ≥ N + 2 p+1\ge N+2 p + 1 ≥ N + 2 . The support of b p + 1 ( 2 + ) b_{p+1}^{(2+)} b p + 1 ( 2 + ) given by (J.7) is a subset of the above support. Hence,
supp [ source term ] ⊆ { N + r N + ∣ r ∈ Z ≥ 1 } . (338) \operatorname{supp}[\text{source term}]\subseteq\{N+rN_{+}\mid r\in\mathbb{Z}_{\ge1}\}. \tag*{(338)} supp [ source term ] ⊆ { N + r N + ∣ r ∈ Z ≥ 1 } . (338)
We next consider the propagator term in eq. (J.45). According to the selection rule (J.18), K p ; m , l , s ( 0 ) ≠ 0 K_{p;m,l,s}^{(0)}\ne0 K p ; m , l , s ( 0 ) = 0 requires
m = l = p , p = s + ( j + 1 ) N + , j ∈ Z ≥ 0 . (339) m=l=p,\qquad p=s+(j+1)N_{+},\qquad j\in\mathbb{Z}_{\ge0}. \tag*{(339)} m = l = p , p = s + ( j + 1 ) N + , j ∈ Z ≥ 0 . (339)
Therefore, every nonzero contribution from G s + 1 ( 2 ) G^{(2)}_{s+1} G s + 1 ( 2 ) to G p + 1 ( 2 ) G^{(2)}_{p+1} G p + 1 ( 2 ) through the propagator term satisfies
( p + 1 ) − ( s + 1 ) ≡ 0 ( m o d N + ) , (340) (p+1)-(s+1)\equiv0 \pmod{N_{+}}, \tag*{(340)} ( p + 1 ) − ( s + 1 ) ≡ 0 ( mod N + ) , (340)
Furthermore, eq. (J.44) shows that G N ( 2 ) G^{(2)}_{N} G N ( 2 ) is the only nonzero element of G p + 1 ( 2 ) G^{(2)}_{p+1} G p + 1 ( 2 ) in the initial range 1 ≤ p + 1 ≤ N + 1 1\leq p+1\leq N+1 1 ≤ p + 1 ≤ N + 1 , and eq. (J.56) shows that the source term is supported on the residue class { N + r N + ∣ r ∈ Z ≥ 1 } \{N+rN_{+}\mid r\in\mathbb{Z}_{\geq1}\} { N + r N + ∣ r ∈ Z ≥ 1 } , then we can show using eq. (J.45) that
supp [ G ( 2 ) ] ⊆ { N + r N + ∣ r ∈ Z ≥ 0 } . (341) \operatorname{supp}\left[G^{(2)}\right]\subseteq\{N+rN_{+}\mid r\in\mathbb{Z}_{\geq0}\}. \tag*{(341)} supp [ G ( 2 ) ] ⊆ { N + r N + ∣ r ∈ Z ≥ 0 } . (341)
Specifically, we prove by induction on r r r the statement P ( r ) P(r) P ( r ) : for 1 ≤ p + 1 ≤ N + r N + 1\leq p+1\leq N+rN_{+} 1 ≤ p + 1 ≤ N + r N + , G p + 1 ( 2 ) G^{(2)}_{p+1} G p + 1 ( 2 ) has support on p + 1 ≡ N ( m o d N + ) p+1\equiv N\pmod{N_{+}} p + 1 ≡ N ( mod N + ) . From eq. (J.44) we see that P ( 0 ) P(0) P ( 0 ) is true and G N + 1 ( 2 ) = 0 G^{(2)}_{N+1}=0 G N + 1 ( 2 ) = 0 , which together with eq. (J.56) shows that P ( 1 ) P(1) P ( 1 ) is also true. Suppose that P ( r ) P(r) P ( r ) holds. We consider the index range for P ( r + 1 ) P(r+1) P ( r + 1 ) with r ≥ 1 r\geq1 r ≥ 1 :
N + r N + + 1 ≤ p + 1 ≤ N + ( r + 1 ) N + . (342) N+rN_{+}+1\leq p+1\leq N+(r+1)N_{+}. \tag*{(342)} N + r N + + 1 ≤ p + 1 ≤ N + ( r + 1 ) N + . (342)
We prove P ( r + 1 ) P(r+1) P ( r + 1 ) by showing that for this index range, G p + 1 ( 2 ) G^{(2)}_{p+1} G p + 1 ( 2 ) can be nonzero only if p + 1 = N + ( r + 1 ) N + p+1=N+(r+1)N_{+} p + 1 = N + ( r + 1 ) N + . First, we note all the G s + 1 ( 2 ) G^{(2)}_{s+1} G s + 1 ( 2 ) entering the propagator term in eq. (J.45) with p + 1 p+1 p + 1 in the range (J.60) must satisfy $1\leq s+1\leq N+rN_{+}$, therefore covered by the induction hypothesis $P(r)$. Thus, nonzero $G^{(2)}_{s+1}$ terms in the propagator term satisfy $s+1\equiv N\pmod{N_{+}}$. For $p+1$ in the range (J.60), the selection rule {xref}(J.58) further implies that the propagator term can be nonzero only when $p+1=N+(r+1)N_{+}$; moreover, according to eq. {xref}(J.56), the source term in {xref}(J.45) can be nonzero only at $p+1=N+(r+1)N_{+}$. The recursion relation {xref}(J.45) then implies that the statement $P(r+1)$ is true. Hence, $P(r)$ is true for any $r\in\mathbb{Z}_{\geq 0}$ by induction, proving eq. {xref}(J.59)`.
Analogous to the recursion relation reduction in appendix J.2, the sparse support (J.59) motivates us to introduce a nonzero subsequence of G p ( 2 ) G^{(2)}_{p} G p ( 2 ) :
g r ( 2 ) ≡ G N + r N + ( 2 ) , r ≥ 0. (343) g^{(2)}_{r}\equiv G^{(2)}_{N+rN_{+}},\qquad r\geq0. \tag*{(343)} g r ( 2 ) ≡ G N + r N + ( 2 ) , r ≥ 0. (343)
It follows from eq. (J.43) that
T 1 ( 2 ) = − ∑ m = 1 ∞ G m ( 2 ) = − ∑ m ∈ supp [ G ( 2 ) ] G m ( 2 ) = − ∑ r = 0 ∞ g r ( 2 ) . (344) T^{(2)}_{1}=-\sum_{m=1}^{\infty}G^{(2)}_{m}=-\sum_{m\in\operatorname{supp}[G^{(2)}]}G^{(2)}_{m}=-\sum_{r=0}^{\infty}g^{(2)}_{r}. \tag*{(344)} T 1 ( 2 ) = − m = 1 ∑ ∞ G m ( 2 ) = − m ∈ supp [ G ( 2 ) ] ∑ G m ( 2 ) = − r = 0 ∑ ∞ g r ( 2 ) . (344)
We therefore introduce the partial sum series
Y r ( 2 ) ≡ ∑ t = 0 r g t ( 2 ) , r ≥ 0 (345) Y^{(2)}_{r}\equiv\sum_{t=0}^{r}g^{(2)}_{t},\qquad r\geq0 \tag*{(345)} Y r ( 2 ) ≡ t = 0 ∑ r g t ( 2 ) , r ≥ 0 (345)
such that
T 1 ( 2 ) = − lim r → ∞ Y r ( 2 ) . (346) T^{(2)}_{1}=-\lim_{r\to\infty}Y^{(2)}_{r}. \tag*{(346)} T 1 ( 2 ) = − r → ∞ lim Y r ( 2 ) . (346)
Next, we translate the recursion relation (J.11) of G p ( 2 ) G^{(2)}_{p} G p ( 2 ) to that of g r ( 2 ) g^{(2)}_{r} g r ( 2 ) to solve the partial sum series Y r ( 2 ) Y^{(2)}_{r} Y r ( 2 ) , which will in turn yield the closed form expression for T 1 ( 2 ) T^{(2)}_{1} T 1 ( 2 ) by taking the r → ∞ r\to\infty r → ∞ limit. To start, we note the zeroth term g 0 ( 2 ) g_{0}^{(2)} g 0 ( 2 ) is given by eq. (J.44) as
g 0 ( 2 ) = G N ( 2 ) = b N ( 2 + ) S ˉ 2 N N + + 1 . (347) g_{0}^{(2)}=G_{N}^{(2)}=b_{N}^{(2+)}\bar{S}_{\frac{2N}{N_{+}+1}}. \tag*{(347)} g 0 ( 2 ) = G N ( 2 ) = b N ( 2 + ) S ˉ N + + 1 2 N . (347)
The remaining terms in g r ( 2 ) g_{r}^{(2)} g r ( 2 ) satisfy r ≥ 1 r\ge1 r ≥ 1 , which correspond to G p + 1 ( 2 ) G_{p+1}^{(2)} G p + 1 ( 2 ) terms with p ≥ 2 N p\ge2N p ≥ 2 N , which are covered by the recursion relation (J.45). To reduce eq. (J.45) to a recursion relation of g r ( 2 ) g_{r}^{(2)} g r ( 2 ) , we set p + 1 = N + r N + p+1=N+rN_{+} p + 1 = N + r N + , s + 1 = 1 + t N + s+1=1+tN_{+} s + 1 = 1 + t N + in the source term and s + 1 = N + t N + s+1=N+tN_{+} s + 1 = N + t N + in the propagator term, which leads to
g r ( 2 ) = b N + r N + ( 2 + ) S ˉ 1 + r + d − ∑ m = N + 1 p r ∑ l = N + 1 m ∑ t = 0 ⌊ l N + ⌋ − 1 K p r ; m , l , t N + ( 2 ) g t ( 0 ) − ∑ m = N + 1 p r ∑ l = N + 1 m ∑ t = 0 ⌊ l − N + 1 N + ⌋ − 1 K p r ; p r , l , N − 1 + t N + ( 0 ) g t ( 2 ) , r ≥ 1 , (348) \begin{aligned}
g_{r}^{(2)}={}&b_{N+rN_{+}}^{(2+)}\bar{S}_{1+r+d}
-\sum_{m=N+1}^{p_{r}}\sum_{l=N+1}^{m}\sum_{t=0}^{\left\lfloor\frac{l}{N_{+}}\right\rfloor-1}K_{p_{r};m,l,tN_{+}}^{(2)}g_{t}^{(0)}\\
&-\sum_{m=N+1}^{p_{r}}\sum_{l=N+1}^{m}\sum_{t=0}^{\left\lfloor\frac{l-N+1}{N_{+}}\right\rfloor-1}K_{p_{r};p_{r},l,N-1+tN_{+}}^{(0)}g_{t}^{(2)},\qquad r\ge1, \tag*{(348)}
\end{aligned} g r ( 2 ) = b N + r N + ( 2 + ) S ˉ 1 + r + d − m = N + 1 ∑ p r l = N + 1 ∑ m t = 0 ∑ ⌊ N + l ⌋ − 1 K p r ; m , l , t N + ( 2 ) g t ( 0 ) − m = N + 1 ∑ p r l = N + 1 ∑ m t = 0 ∑ ⌊ N + l − N + 1 ⌋ − 1 K p r ; p r , l , N − 1 + t N + ( 0 ) g t ( 2 ) , r ≥ 1 , (348)
where we have used eqs. (J.20) and (J.61) and
p r ≡ N + r N + − 1. (349) p_{r}\equiv N+rN_{+}-1. \tag*{(349)} p r ≡ N + r N + − 1. (349)
To simplify eq. (J.66), we first reduce the propagator term using the selection rule (J.18) imposed by K ( 0 ) K^{(0)} K ( 0 ) , which gives
propagator term = ∑ t = 0 ⌊ p r − N + 1 N + ⌋ − 1 K p r ; p r , p r , N − 1 + t N + ( 0 ) g t ( 2 ) = a 2 ( r + d ) 2 ∑ t = 0 r − 1 g t ( 2 ) , r ≥ 1. (350) \operatorname{propagator\ term}=\sum_{t=0}^{\left\lfloor\frac{p_{r}-N+1}{N_{+}}\right\rfloor-1}K_{p_{r};p_{r},p_{r},N-1+tN_{+}}^{(0)}g_{t}^{(2)}=\frac{a^{2}}{(r+d)^{2}}\sum_{t=0}^{r-1}g_{t}^{(2)},\qquad r\ge1. \tag*{(350)} propagator term = t = 0 ∑ ⌊ N + p r − N + 1 ⌋ − 1 K p r ; p r , p r , N − 1 + t N + ( 0 ) g t ( 2 ) = ( r + d ) 2 a 2 t = 0 ∑ r − 1 g t ( 2 ) , r ≥ 1. (350)
where we have used eqs. (J.7) and (J.10), and where a a a and d d d are indices defined in eq. (J.5). To reduce the source term in eq. (J.66), we use eq. (J.47) and the selection rules (J.51) and (J.52) to obtain
source term = b N + r N + ( 2 + ) S ˉ 1 + r + d − ∑ l = N + 1 p r ∑ t = 0 ⌊ l N + ⌋ − 1 ( K p r ; l , l , t N + ( 2 + ) + K p r ; p r , l , t N + ( 2 − ) ) g t ( 0 ) = b N + r N + ( 2 + ) S ˉ 1 + r + d − ∑ l = N + p r ( b p r − l + 1 ( 2 + ) l + b p r − l + 1 ( 2 − ) p r ) N S ˉ 1 + p r N + 2 l S ˉ 1 + l N + ∑ t = 0 ⌊ l N + ⌋ − 1 δ ( l N + − t − 1 ) g t ( 0 ) = b N + r N + ( 2 + ) S ˉ 1 + r + d − ∑ q = 1 ⌊ p r N + ⌋ ( b p r − q N + + 1 ( 2 + ) q N + + b p r − q N + + 1 ( 2 − ) p r ) N 2 q N + S ˉ 1 + p r N + S ˉ 1 + q Y q − 1 ( 0 ) , r ≥ 1 , (351) \begin{aligned}
\operatorname{source\ term}={}&b_{N+rN_{+}}^{(2+)}\bar{S}_{1+r+d}
-\sum_{l=N+1}^{p_{r}}\sum_{t=0}^{\left\lfloor\frac{l}{N_{+}}\right\rfloor-1}
\left(K_{p_{r};l,l,tN_{+}}^{(2+)}+K_{p_{r};p_{r},l,tN_{+}}^{(2-)}\right)g_{t}^{(0)}\\
={}&b_{N+rN_{+}}^{(2+)}\bar{S}_{1+r+d}
-\sum_{l=N_{+}}^{p_{r}}\left(\frac{b_{p_{r}-l+1}^{(2+)}}{l}+\frac{b_{p_{r}-l+1}^{(2-)}}{p_{r}}\right)
\frac{N\bar{S}_{1+\frac{p_{r}}{N_{+}}}}{2l\bar{S}_{1+\frac{l}{N_{+}}}}
\sum_{t=0}^{\left\lfloor\frac{l}{N_{+}}\right\rfloor-1}
\delta\left(\frac{l}{N_{+}}-t-1\right)g_{t}^{(0)}\\
={}&b_{N+rN_{+}}^{(2+)}\bar{S}_{1+r+d}
-\sum_{q=1}^{\left\lfloor\frac{p_{r}}{N_{+}}\right\rfloor}
\left(\frac{b_{p_{r}-qN_{+}+1}^{(2+)}}{qN_{+}}+\frac{b_{p_{r}-qN_{+}+1}^{(2-)}}{p_{r}}\right)
\frac{N}{2qN_{+}}\frac{\bar{S}_{1+\frac{p_{r}}{N_{+}}}}{\bar{S}_{1+q}}Y_{q-1}^{(0)},\qquad r\ge1, \tag*{(351)}
\end{aligned} source term = = = b N + r N + ( 2 + ) S ˉ 1 + r + d − l = N + 1 ∑ p r t = 0 ∑ ⌊ N + l ⌋ − 1 ( K p r ; l , l , t N + ( 2 + ) + K p r ; p r , l , t N + ( 2 − ) ) g t ( 0 ) b N + r N + ( 2 + ) S ˉ 1 + r + d − l = N + ∑ p r ( l b p r − l + 1 ( 2 + ) + p r b p r − l + 1 ( 2 − ) ) 2 l S ˉ 1 + N + l N S ˉ 1 + N + p r t = 0 ∑ ⌊ N + l ⌋ − 1 δ ( N + l − t − 1 ) g t ( 0 ) b N + r N + ( 2 + ) S ˉ 1 + r + d − q = 1 ∑ ⌊ N + p r ⌋ ( q N + b p r − q N + + 1 ( 2 + ) + p r b p r − q N + + 1 ( 2 − ) ) 2 q N + N S ˉ 1 + q S ˉ 1 + N + p r Y q − 1 ( 0 ) , r ≥ 1 , (351)
where in arriving at the last line we used l = q N + l=qN_{+} l = q N + with q ∈ Z ≥ 1 q\in\mathbb{Z}_{\geq1} q ∈ Z ≥ 1 imposed by the delta factor and the partial sum Y q − 1 ( 0 ) Y^{(0)}_{q-1} Y q − 1 ( 0 ) defined in eq. (J.22). The definition (J.67) of p r p_r p r implies ⌊ p r / N + ⌋ = r \lfloor p_r/N_{+}\rfloor=r ⌊ p r / N + ⌋ = r . This, with the index notations (J.5), further reduces eq. (J.69) to
source term = b N + r N + ( 2 + ) S ‾ 1 + r + d − N 2 N + 2 ∑ q = 1 r ( b N + ( r − q ) N + ( 2 + ) q + b N + ( r − q ) N + ( 2 − ) r + d ) 1 q S ‾ 1 + r + d S ‾ 1 + q Y q − 1 ( 0 ) , r ≥ 1. (352) \text{source term}=b^{(2+)}_{N+rN_{+}}\overline{S}_{1+r+d}-\frac{N}{2N_{+}^{2}}\sum_{q=1}^{r}\left(\frac{b^{(2+)}_{N+(r-q)N_{+}}}{q}+\frac{b^{(2-)}_{N+(r-q)N_{+}}}{r+d}\right)\frac{1}{q}\frac{\overline{S}_{1+r+d}}{\overline{S}_{1+q}}Y^{(0)}_{q-1},\quad r\geq1. \tag*{(352)} source term = b N + r N + ( 2 + ) S 1 + r + d − 2 N + 2 N q = 1 ∑ r q b N + ( r − q ) N + ( 2 + ) + r + d b N + ( r − q ) N + ( 2 − ) q 1 S 1 + q S 1 + r + d Y q − 1 ( 0 ) , r ≥ 1. (352)
For r = 1 r=1 r = 1 , the above becomes
source term = b 2 N + + 1 ( 2 + ) S ‾ 2 + d − N 2 N + 2 ( b N ( 2 + ) + b N ( 2 − ) 1 + d ) S ‾ 2 + d S ‾ 2 Y 0 ( 0 ) , r = 1. (353) \text{source term}=b^{(2+)}_{2N_{+}+1}\overline{S}_{2+d}-\frac{N}{2N_{+}^{2}}\left(b^{(2+)}_{N}+\frac{b^{(2-)}_{N}}{1+d}\right)\frac{\overline{S}_{2+d}}{\overline{S}_{2}}Y^{(0)}_{0},\quad r=1. \tag*{(353)} source term = b 2 N + + 1 ( 2 + ) S 2 + d − 2 N + 2 N ( b N ( 2 + ) + 1 + d b N ( 2 − ) ) S 2 S 2 + d Y 0 ( 0 ) , r = 1. (353)
For r ≥ 2 r\geq2 r ≥ 2 , the support of b ( 2 ± ) b^{(2\pm)} b ( 2 ± ) given in eq. (J.8) requires q ∈ { r , r − 1 } q\in\{r,r-1\} q ∈ { r , r − 1 } in the summation in eq. (J.70), which gives
source term = − N 2 N + 2 ( b N ( 2 + ) r + b N ( 2 − ) r + d ) 1 r S ‾ 1 + r + d S ‾ 1 + r Y r − 1 ( 0 ) − N 2 N + 2 ( b 2 N + + 1 ( 2 + ) r − 1 + b 2 N + + 1 ( 2 − ) r + d ) 1 r − 1 S ‾ 1 + r + d S ‾ r Y r − 2 ( 0 ) , r ≥ 2. (354) \begin{aligned}
\text{source term}={}&-\frac{N}{2N_{+}^{2}}\left(\frac{b^{(2+)}_{N}}{r}+\frac{b^{(2-)}_{N}}{r+d}\right)\frac{1}{r}\frac{\overline{S}_{1+r+d}}{\overline{S}_{1+r}}Y^{(0)}_{r-1}\\
&-\frac{N}{2N_{+}^{2}}\left(\frac{b^{(2+)}_{2N_{+}+1}}{r-1}+\frac{b^{(2-)}_{2N_{+}+1}}{r+d}\right)\frac{1}{r-1}\frac{\overline{S}_{1+r+d}}{\overline{S}_{r}}Y^{(0)}_{r-2},\quad r\geq2. \tag*{(354)}
\end{aligned} source term = − 2 N + 2 N ( r b N ( 2 + ) + r + d b N ( 2 − ) ) r 1 S 1 + r S 1 + r + d Y r − 1 ( 0 ) − 2 N + 2 N ( r − 1 b 2 N + + 1 ( 2 + ) + r + d b 2 N + + 1 ( 2 − ) ) r − 1 1 S r S 1 + r + d Y r − 2 ( 0 ) , r ≥ 2. (354)
where we have used b N + r N + ( 2 + ) = 0 b^{(2+)}_{N+rN_{+}}=0 b N + r N + ( 2 + ) = 0 for r ≥ 2 r\geq2 r ≥ 2 . Combining the expression (J.65) for the first term g 0 ( 2 ) g^{(2)}_{0} g 0 ( 2 ) , the propagator term expression (J.68), the source term expressions (J.71) and (J.72), and the expression (J.7) for b ( 0 ) b^{(0)} b ( 0 ) and b ( 2 ± ) b^{(2\pm)} b ( 2 ± ) , we reduce the recursion relation (J.66) for g r ( 2 ) g^{(2)}_{r} g r ( 2 ) to
g r ( 2 ) = { i a d S ‾ 1 + d , r = 0 , − N 2 r S ‾ 2 + d − 1 16 i a d S ‾ 1 + d − S ‾ 2 + d S ‾ 2 π a 2 2 N + , r = 1 , J r ( 2 ) − a 2 ( r + d ) 2 ∑ t = 0 r − 1 g t ( 2 ) , r ≥ 2. (355) g^{(2)}_{r}=
\begin{cases}
\dfrac{ia}{d}\overline{S}_{1+d}, & r=0,\\
-\dfrac{N}{2r}\overline{S}_{2+d}-\dfrac{1}{16}\dfrac{ia}{d}\overline{S}_{1+d}-\dfrac{\overline{S}_{2+d}}{\overline{S}_{2}}\dfrac{\pi a^{2}}{2N_{+}}, & r=1,\\
J^{(2)}_{r}-\dfrac{a^{2}}{(r+d)^{2}}\sum_{t=0}^{r-1}g^{(2)}_{t}, & r\geq2.
\end{cases} \tag*{(355)} g r ( 2 ) = ⎩ ⎨ ⎧ d ia S 1 + d , − 2 r N S 2 + d − 16 1 d ia S 1 + d − S 2 S 2 + d 2 N + π a 2 , J r ( 2 ) − ( r + d ) 2 a 2 ∑ t = 0 r − 1 g t ( 2 ) , r = 0 , r = 1 , r ≥ 2. (355)
where
J r ( 2 ) = − i a 2 N + r 2 ( r + d ) S ‾ 1 + r + d S ‾ 1 + r Y r − 1 ( 0 ) + a 2 ( 2 r + d − 1 ) ( r − 1 ) 2 ( r + d ) S ‾ 1 + r + d S ‾ r Y r − 2 ( 0 ) , r ≥ 2. (356) J^{(2)}_{r}=-\frac{ia^{2}}{N_{+}r^{2}(r+d)}\frac{\overline{S}_{1+r+d}}{\overline{S}_{1+r}}Y^{(0)}_{r-1}+\frac{a^{2}(2r+d-1)}{(r-1)^{2}(r+d)}\frac{\overline{S}_{1+r+d}}{\overline{S}_{r}}Y^{(0)}_{r-2},\quad r\geq2. \tag*{(356)} J r ( 2 ) = − N + r 2 ( r + d ) i a 2 S 1 + r S 1 + r + d Y r − 1 ( 0 ) + ( r − 1 ) 2 ( r + d ) a 2 ( 2 r + d − 1 ) S r S 1 + r + d Y r − 2 ( 0 ) , r ≥ 2. (356)
Recall that we have defined in eq. (J.63) the partial sum Y r ( 2 ) Y^{(2)}_{r} Y r ( 2 ) of g t ( 2 ) g^{(2)}_{t} g t ( 2 ) . The recursion relation (J.73) then translates to
Y r ( 2 ) = { g 0 ( 2 ) = i a d S ‾ 1 + d , r = 0 , g 0 ( 2 ) + g 1 ( 2 ) = − N 2 S ‾ 2 + d + 15 16 i a d S ‾ 1 + d − S ‾ 2 + d S ‾ 2 π a 2 2 N + , r = 1 , [ 1 − a 2 ( r + d ) 2 ] Y r − 1 ( 2 ) + J r ( 2 ) , r ≥ 2. (357) Y^{(2)}_{r}=
\begin{cases}
g^{(2)}_{0}=\dfrac{ia}{d}\overline{S}_{1+d}, & r=0,\\
g^{(2)}_{0}+g^{(2)}_{1}=-\dfrac{N}{2}\overline{S}_{2+d}+\dfrac{15}{16}\dfrac{ia}{d}\overline{S}_{1+d}-\dfrac{\overline{S}_{2+d}}{\overline{S}_{2}}\dfrac{\pi a^{2}}{2N_{+}}, & r=1,\\
\left[1-\dfrac{a^{2}}{(r+d)^{2}}\right]Y^{(2)}_{r-1}+J^{(2)}_{r}, & r\geq2.
\end{cases} \tag*{(357)} Y r ( 2 ) = ⎩ ⎨ ⎧ g 0 ( 2 ) = d ia S 1 + d , g 0 ( 2 ) + g 1 ( 2 ) = − 2 N S 2 + d + 16 15 d ia S 1 + d − S 2 S 2 + d 2 N + π a 2 , [ 1 − ( r + d ) 2 a 2 ] Y r − 1 ( 2 ) + J r ( 2 ) , r = 0 , r = 1 , r ≥ 2. (357)
where we have used the property g r ( 2 ) = Y r ( 2 ) − Y r − 1 ( 2 ) g^{(2)}_{r}=Y^{(2)}_{r}-Y^{(2)}_{r-1} g r ( 2 ) = Y r ( 2 ) − Y r − 1 ( 2 ) for r ≥ 1 r\geq1 r ≥ 1 . Iterating the third line of eq. (J.75) we obtain the expression
Y r ( 2 ) = ( ∏ j = 2 r [ 1 − a 2 ( j + d ) 2 ] ) Y 1 ( 2 ) + ∑ q = 2 r ( ∏ j = q + 1 r [ 1 − a 2 ( j + d ) 2 ] ) J q ( 2 ) , r ≥ 2. (358) Y^{(2)}_{r}=\left(\prod_{j=2}^{r}\left[1-\frac{a^{2}}{(j+d)^{2}}\right]\right)Y^{(2)}_{1}+\sum_{q=2}^{r}\left(\prod_{j=q+1}^{r}\left[1-\frac{a^{2}}{(j+d)^{2}}\right]\right)J^{(2)}_{q},\quad r\geq2. \tag*{(358)} Y r ( 2 ) = ( j = 2 ∏ r [ 1 − ( j + d ) 2 a 2 ] ) Y 1 ( 2 ) + q = 2 ∑ r ( j = q + 1 ∏ r [ 1 − ( j + d ) 2 a 2 ] ) J q ( 2 ) , r ≥ 2. (358)
which is analogous to the product expression (J.26) for Y r ( 0 ) Y_r^{(0)} Y r ( 0 ) but with an additional source term. Note the product factor is expressible in terms of gamma functions
∏ j = q + 1 r [ 1 − a 2 ( j + d ) 2 ] = [ ∏ j = q + 1 r ( j + d + a j + d ) ] [ ∏ j = q + 1 r ( j + d − a j + d ) ] = [ Γ ( q + 1 + d ) ] 2 [ Γ ( r + 1 + d ) ] 2 Γ ( r + 1 + d + a ) Γ ( r + 1 + d − a ) Γ ( q + 1 + d + a ) Γ ( q + 1 + d − a ) . (359) \prod_{j=q+1}^{r}\left[1-\frac{a^2}{(j+d)^2}\right]
=
\left[\prod_{j=q+1}^{r}\left(\frac{j+d+a}{j+d}\right)\right]
\left[\prod_{j=q+1}^{r}\left(\frac{j+d-a}{j+d}\right)\right]
=
\frac{\left[\Gamma(q+1+d)\right]^2}{\left[\Gamma(r+1+d)\right]^2}
\frac{\Gamma(r+1+d+a)\Gamma(r+1+d-a)}
{\Gamma(q+1+d+a)\Gamma(q+1+d-a)} . \tag*{(359)} j = q + 1 ∏ r [ 1 − ( j + d ) 2 a 2 ] = [ j = q + 1 ∏ r ( j + d j + d + a ) ] [ j = q + 1 ∏ r ( j + d j + d − a ) ] = [ Γ ( r + 1 + d ) ] 2 [ Γ ( q + 1 + d ) ] 2 Γ ( q + 1 + d + a ) Γ ( q + 1 + d − a ) Γ ( r + 1 + d + a ) Γ ( r + 1 + d − a ) . (359)
In the r → ∞ r \to\infty r → ∞ limit, we find
lim r → ∞ ∏ j = q + 1 r [ 1 − a 2 ( j + d ) 2 ] = [ Γ ( q + 1 + d ) ] 2 Γ ( q + 1 + d + a ) Γ ( q + 1 + d − a ) . (360) \lim_{r\to\infty}\prod_{j=q+1}^{r}\left[1-\frac{a^2}{(j+d)^2}\right]
=
\frac{\left[\Gamma(q+1+d)\right]^2}
{\Gamma(q+1+d+a)\Gamma(q+1+d-a)} . \tag*{(360)} r → ∞ lim j = q + 1 ∏ r [ 1 − ( j + d ) 2 a 2 ] = Γ ( q + 1 + d + a ) Γ ( q + 1 + d − a ) [ Γ ( q + 1 + d ) ] 2 . (360)
The quadratic coefficient of T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) is therefore given by eqs. (J.64) and (J.76) as
T 1 ( 2 ) = − lim r → ∞ Y r ( 2 ) = − Y 1 ( 2 ) ( 2 + d ) a ( 2 + d ) − a − ∑ r = 2 ∞ J r ( 2 ) ( r + 1 + d ) a ( r + 1 + d ) − a . (361) T_1^{(2)}=-\lim_{r\to\infty}Y_r^{(2)}
=-\frac{Y_1^{(2)}}{(2+d)a(2+d)_{-a}}
-\sum_{r=2}^{\infty}\frac{J_r^{(2)}}{(r+1+d)a(r+1+d)_{-a}} . \tag*{(361)} T 1 ( 2 ) = − r → ∞ lim Y r ( 2 ) = − ( 2 + d ) a ( 2 + d ) − a Y 1 ( 2 ) − r = 2 ∑ ∞ ( r + 1 + d ) a ( r + 1 + d ) − a J r ( 2 ) . (361)
where we have used the Pochhammer symbol to make the expression more compact. The summation term in the above equation can be expanded using the definition (J.74) for J r ( 2 ) J_r^{(2)} J r ( 2 ) and the solution (J.26) for Y r ( 0 ) Y_r^{(0)} Y r ( 0 ) :
∑ r = 2 ∞ J r ( 2 ) ( r + 1 + d ) a ( r + 1 + d ) − a = − ( 2 i N + ) d 4 π a 3 N + ( Σ 1 + 2 Σ 2 ) (362) \sum_{r=2}^{\infty}\frac{J_r^{(2)}}{(r+1+d)a(r+1+d)_{-a}}
=
-\left(\frac{2}{iN_+}\right)^d\frac{4\pi a^3}{N_+}(\Sigma_1+2\Sigma_2) \tag*{(362)} r = 2 ∑ ∞ ( r + 1 + d ) a ( r + 1 + d ) − a J r ( 2 ) = − ( i N + 2 ) d N + 4 π a 3 ( Σ 1 + 2 Σ 2 ) (362)
where we have used eqs. (J.4) and (J.5), and defined
Σ 1 ≡ ∑ r = 2 ∞ 1 ( r + 1 + d ) a ( r + 1 + d ) − a ( r + 1 ) d 1 r 2 ( r + d ) ( 1 + a ) r − 1 ( 1 − a ) r − 1 [ ( r − 1 ) ! ] 2 , (363) \Sigma_1\equiv
\sum_{r=2}^{\infty}
\frac{1}{(r+1+d)a(r+1+d)_{-a}(r+1)_d}
\frac{1}{r^2(r+d)}
\frac{(1+a)_{r-1}(1-a)_{r-1}}{[(r-1)!]^2}, \tag*{(363)} Σ 1 ≡ r = 2 ∑ ∞ ( r + 1 + d ) a ( r + 1 + d ) − a ( r + 1 ) d 1 r 2 ( r + d ) 1 [( r − 1 )! ] 2 ( 1 + a ) r − 1 ( 1 − a ) r − 1 , (363)
Σ 2 ≡ ∑ r = 2 ∞ 1 ( r + 1 + d ) a ( r + 1 + d ) − a ( r + 1 ) d 2 r + d − 1 r ( r − 1 ) 2 ( r + d ) ( 1 + a ) r − 2 ( 1 − a ) r − 2 [ ( r − 2 ) ! ] 2 . (364) \Sigma_2\equiv
\sum_{r=2}^{\infty}
\frac{1}{(r+1+d)a(r+1+d)_{-a}(r+1)_d}
\frac{2r+d-1}{r(r-1)^2(r+d)}
\frac{(1+a)_{r-2}(1-a)_{r-2}}{[(r-2)!]^2}. \tag*{(364)} Σ 2 ≡ r = 2 ∑ ∞ ( r + 1 + d ) a ( r + 1 + d ) − a ( r + 1 ) d 1 r ( r − 1 ) 2 ( r + d ) 2 r + d − 1 [( r − 2 )! ] 2 ( 1 + a ) r − 2 ( 1 − a ) r − 2 . (364)
We next derive closed form expressions for Σ 1 \Sigma_1 Σ 1 and Σ 2 \Sigma_2 Σ 2 , which will yield a closed form expression for T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) through eq. (J.79). The key to reducing Σ 1 \Sigma_1 Σ 1 and Σ 2 \Sigma_2 Σ 2 is to rewrite them in terms of the generalized hypergeometric functions
p F q ( a 1 , … , a p b 1 , … , b q ; z ) ≡ ∑ s = 0 ∞ ( a 1 ) s ⋯ ( a p ) s ( b 1 ) s ⋯ ( b q ) s z s s ! . (365) {}_pF_q\left(\begin{matrix}a_1,\ldots,a_p\\b_1,\ldots,b_q\end{matrix};z\right)
\equiv
\sum_{s=0}^{\infty}
\frac{(a_1)_s\cdots(a_p)_s}{(b_1)_s\cdots(b_q)_s}\frac{z^s}{s!}. \tag*{(365)} p F q ( a 1 , … , a p b 1 , … , b q ; z ) ≡ s = 0 ∑ ∞ ( b 1 ) s ⋯ ( b q ) s ( a 1 ) s ⋯ ( a p ) s s ! z s . (365)
We first shift the summation indices in Σ 1 \Sigma_1 Σ 1 and Σ 2 \Sigma_2 Σ 2 from r r r to s = r − 2 s=r-2 s = r − 2 , giving
Σ 1 = ∑ s = 0 ∞ 1 ( s + 3 + d ) a ( s + 3 + d ) − a ( s + 3 ) d 1 ( s + 2 ) 2 ( s + 2 + d ) ( 1 + a ) s + 1 ( 1 − a ) s + 1 [ ( s + 1 ) ! ] 2 , (366) \Sigma_1=
\sum_{s=0}^{\infty}
\frac{1}{(s+3+d)a(s+3+d)_{-a}(s+3)_d}
\frac{1}{(s+2)^2(s+2+d)}
\frac{(1+a)_{s+1}(1-a)_{s+1}}{[(s+1)!]^2}, \tag*{(366)} Σ 1 = s = 0 ∑ ∞ ( s + 3 + d ) a ( s + 3 + d ) − a ( s + 3 ) d 1 ( s + 2 ) 2 ( s + 2 + d ) 1 [( s + 1 )! ] 2 ( 1 + a ) s + 1 ( 1 − a ) s + 1 , (366)
Σ 2 = ∑ s = 0 ∞ 1 ( s + 3 + d ) a ( s + 3 + d ) − a ( s + 3 ) d 2 s + d + 3 ( s + 2 ) ( s + 1 ) 2 ( s + 2 + d ) ( 1 + a ) s ( 1 − a ) s ( s ! ) 2 . (367) \Sigma_2=
\sum_{s=0}^{\infty}
\frac{1}{(s+3+d)a(s+3+d)_{-a}(s+3)_d}
\frac{2s+d+3}{(s+2)(s+1)^2(s+2+d)}
\frac{(1+a)_s(1-a)_s}{(s!)^2}. \tag*{(367)} Σ 2 = s = 0 ∑ ∞ ( s + 3 + d ) a ( s + 3 + d ) − a ( s + 3 ) d 1 ( s + 2 ) ( s + 1 ) 2 ( s + 2 + d ) 2 s + d + 3 ( s ! ) 2 ( 1 + a ) s ( 1 − a ) s . (367)
We then utilize the gamma function properties
Γ ( x + 1 ) = x Γ ( x ) , s + x = x ( x + 1 ) s ( x ) s (368) \Gamma(x+1)=x\Gamma(x),\qquad s+x=x\frac{(x+1)_s}{(x)_s} \tag*{(368)} Γ ( x + 1 ) = x Γ ( x ) , s + x = x ( x ) s ( x + 1 ) s (368)
to rewrite the summands in Σ 1 \Sigma_1 Σ 1 and Σ 2 \Sigma_2 Σ 2 in the form of the summand in eq. (J.83):
Σ 1 = 1 − a 2 2 Γ ( 2 + d ) Γ ( 3 + d + a ) Γ ( 3 + d − a ) ∑ s = 0 ∞ ( 2 + d ) s ( 2 + a ) s ( 2 − a ) s ( 3 + d + a ) s ( 3 + d − a ) s ( 3 ) s (369) \Sigma_1=\frac{1-a^2}{2}\frac{\Gamma(2+d)}{\Gamma(3+d+a)\Gamma(3+d-a)}\sum_{s=0}^{\infty}\frac{(2+d)_s(2+a)_s(2-a)_s}{(3+d+a)_s(3+d-a)_s(3)_s} \tag*{(369)} Σ 1 = 2 1 − a 2 Γ ( 3 + d + a ) Γ ( 3 + d − a ) Γ ( 2 + d ) s = 0 ∑ ∞ ( 3 + d + a ) s ( 3 + d − a ) s ( 3 ) s ( 2 + d ) s ( 2 + a ) s ( 2 − a ) s (369)
Σ 2 = 2 Γ ( 2 ) Γ ( 2 + d ) Γ ( 3 + d + a ) Γ ( 3 + d − a ) Γ ( d + 5 2 ) Γ ( 3 + d 2 ) ∑ s = 0 ∞ ( 1 + a ) s ( 1 − a ) s ( 1 ) s ( d + 5 2 ) s ( 2 + d ) s ( 3 + d + a ) s ( 3 + d − a ) s ( 2 ) s ( 3 + d 2 ) s 1 s ! . (370) \Sigma_2=\frac{2\Gamma(2)\Gamma(2+d)}{\Gamma(3+d+a)\Gamma(3+d-a)}\frac{\Gamma\left(\frac{d+5}{2}\right)}{\Gamma\left(\frac{3+d}{2}\right)}\sum_{s=0}^{\infty}\frac{(1+a)_s(1-a)_s(1)_s\left(\frac{d+5}{2}\right)_s(2+d)_s}{(3+d+a)_s(3+d-a)_s(2)_s\left(\frac{3+d}{2}\right)_s}\frac{1}{s!}. \tag*{(370)} Σ 2 = Γ ( 3 + d + a ) Γ ( 3 + d − a ) 2Γ ( 2 ) Γ ( 2 + d ) Γ ( 2 3 + d ) Γ ( 2 d + 5 ) s = 0 ∑ ∞ ( 3 + d + a ) s ( 3 + d − a ) s ( 2 ) s ( 2 3 + d ) s ( 1 + a ) s ( 1 − a ) s ( 1 ) s ( 2 d + 5 ) s ( 2 + d ) s s ! 1 . (370)
Noting s ! = ( 1 ) s s!=(1)_s s ! = ( 1 ) s and d = 4 a − 1 d=4a-1 d = 4 a − 1 , we express the above in terms of the generalized hypergeometric functions
Σ 1 = ( 1 − a 2 ) Γ ( 4 a + 1 ) 2 Γ ( 5 a + 2 ) Γ ( 3 a + 2 ) F 4 , Σ 2 = 2 ( 2 a + 1 ) Γ ( 4 a + 1 ) Γ ( 5 a + 2 ) Γ ( 3 a + 2 ) F 5 (371) \Sigma_1=\frac{(1-a^2)\Gamma(4a+1)}{2\Gamma(5a+2)\Gamma(3a+2)}F_4,\qquad\Sigma_2=\frac{2(2a+1)\Gamma(4a+1)}{\Gamma(5a+2)\Gamma(3a+2)}F_5 \tag*{(371)} Σ 1 = 2Γ ( 5 a + 2 ) Γ ( 3 a + 2 ) ( 1 − a 2 ) Γ ( 4 a + 1 ) F 4 , Σ 2 = Γ ( 5 a + 2 ) Γ ( 3 a + 2 ) 2 ( 2 a + 1 ) Γ ( 4 a + 1 ) F 5 (371)
F 4 ≡ 4 F 3 ( 4 a + 1 , 2 + a , 2 − a , 1 5 a + 2 , 3 a + 2 , 3 ; z = 1 ) . (372) F_4\equiv{}_4F_3\left(\begin{matrix}4a+1,\ 2+a,\ 2-a,\ 1\\5a+2,\ 3a+2,\ 3\end{matrix};z=1\right). \tag*{(372)} F 4 ≡ 4 F 3 ( 4 a + 1 , 2 + a , 2 − a , 1 5 a + 2 , 3 a + 2 , 3 ; z = 1 ) . (372)
F 5 ≡ 5 F 4 ( 1 + a , 1 − a , 1 , 2 a + 2 , 4 a + 1 5 a + 2 , 3 a + 2 , 2 , 2 a + 1 ; z = 1 ) . (373) F_5\equiv{}_5F_4\left(\begin{matrix}1+a,\ 1-a,\ 1,\ 2a+2,\ 4a+1\\5a+2,\ 3a+2,\ 2,\ 2a+1\end{matrix};z=1\right). \tag*{(373)} F 5 ≡ 5 F 4 ( 1 + a , 1 − a , 1 , 2 a + 2 , 4 a + 1 5 a + 2 , 3 a + 2 , 2 , 2 a + 1 ; z = 1 ) . (373)
where F 4 F_4 F 4 can be simplified using Mathematica
F 4 = 2 ( 1 + 5 a ) ( 1 − a ) ( 1 + a ) ( 1 − 4 a ) ( − 11 − 4 a + 3 a + ( 4 1 − 4 a ) ( 5 π ) Γ ( 5 a ) Γ ( 2 + 3 a ) Γ ( 1 2 + a ) [ Γ ( 1 + 2 a ) ] 2 Γ ( 1 2 + 3 a ) ) (374) F_4=\frac{2(1+5a)}{(1-a)(1+a)(1-4a)}\left(-11-\frac{4}{a}+3a+\frac{(4^{1-4a})(5\pi)\Gamma(5a)\Gamma(2+3a)}{\Gamma\left(\frac{1}{2}+a\right)[\Gamma(1+2a)]^2\Gamma\left(\frac{1}{2}+3a\right)}\right) \tag*{(374)} F 4 = ( 1 − a ) ( 1 + a ) ( 1 − 4 a ) 2 ( 1 + 5 a ) ( − 11 − a 4 + 3 a + Γ ( 2 1 + a ) [ Γ ( 1 + 2 a ) ] 2 Γ ( 2 1 + 3 a ) ( 4 1 − 4 a ) ( 5 π ) Γ ( 5 a ) Γ ( 2 + 3 a ) ) (374)
whereas F 5 F_5 F 5 cannot be simplified so. To evaluate F 5 F_5 F 5 , we use Dixon’s theorem (see e.g. [38 ] )
D ( A , B , C ) ≡ 3 F 2 ( A , B , C 1 + A − B , 1 + A − C ; z = 1 ) = ( 1 + A ) − B ( 1 + A 2 − C ) − B ( 1 + A 2 ) − B ( 1 + A − C ) − B (375) \mathcal{D}(A,B,C)\equiv{}_3F_2\left(\begin{matrix}A,\ B,\ C\\1+A-B,\ 1+A-C\end{matrix};z=1\right)=\frac{(1+A)_{-B}\left(1+\frac{A}{2}-C\right)_{-B}}{\left(1+\frac{A}{2}\right)_{-B}(1+A-C)_{-B}} \tag*{(375)} D ( A , B , C ) ≡ 3 F 2 ( A , B , C 1 + A − B , 1 + A − C ; z = 1 ) = ( 1 + 2 A ) − B ( 1 + A − C ) − B ( 1 + A ) − B ( 1 + 2 A − C ) − B (375)
valid for ℜ ( 1 + A / 2 − B − C ) > 0 \Re(1+A/2-B-C)>0 ℜ ( 1 + A /2 − B − C ) > 0 . Here ( A , B , C ) (A,B,C) ( A , B , C ) are auxiliary complex parameters, distinct from the quantities denoted by the same symbols in the main text. We consider
F 5 = ∑ s = 0 ∞ u s , D ( 4 a + 1 , 1 + a , 1 − a ) = ∑ s = 0 ∞ v s , (376) F_5=\sum_{s=0}^{\infty}u_s,\qquad\mathcal{D}(4a+1,1+a,1-a)=\sum_{s=0}^{\infty}v_s, \tag*{(376)} F 5 = s = 0 ∑ ∞ u s , D ( 4 a + 1 , 1 + a , 1 − a ) = s = 0 ∑ ∞ v s , (376)
where
u s ≡ ( 1 + a ) s ( 1 − a ) s ( 2 a + 2 ) s ( 4 a + 1 ) s ( 5 a + 2 ) s ( 3 a + 2 ) s ( 2 ) s ( 2 a + 1 ) s , u_s\equiv\frac{(1+a)_s(1-a)_s(2a+2)_s(4a+1)_s}{(5a+2)_s(3a+2)_s(2)_s(2a+1)_s}, u s ≡ ( 5 a + 2 ) s ( 3 a + 2 ) s ( 2 ) s ( 2 a + 1 ) s ( 1 + a ) s ( 1 − a ) s ( 2 a + 2 ) s ( 4 a + 1 ) s ,
v s ≡ ( 4 a + 1 ) s ( 1 + a ) s ( 1 − a ) s ( 3 a + 1 ) s ( 5 a + 1 ) s s ! . (377) v_s\equiv\frac{(4a+1)_s(1+a)_s(1-a)_s}{(3a+1)_s(5a+1)_s\,s!}. \tag*{(377)} v s ≡ ( 3 a + 1 ) s ( 5 a + 1 ) s s ! ( 4 a + 1 ) s ( 1 + a ) s ( 1 − a ) s . (377)
Using eq. (J.86) we obtain
u s v s = ( 3 a + 1 ) ( 5 a + 1 ) 2 a + 1 s + 2 a + 1 ( s + 3 a + 1 ) ( s + 5 a + 1 ) ( s + 1 ) (378) \frac{u_s}{v_s}=\frac{(3a+1)(5a+1)}{2a+1}\frac{s+2a+1}{(s+3a+1)(s+5a+1)(s+1)} \tag*{(378)} v s u s = 2 a + 1 ( 3 a + 1 ) ( 5 a + 1 ) ( s + 3 a + 1 ) ( s + 5 a + 1 ) ( s + 1 ) s + 2 a + 1 (378)
and
v s + 1 v s = ( s + 4 a + 1 ) ( s + a + 1 ) ( s − a + 1 ) ( s + 3 a + 1 ) ( s + 5 a + 1 ) ( s + 1 ) . (379) \frac{v_{s+1}}{v_s}=\frac{(s+4a+1)(s+a+1)(s-a+1)}{(s+3a+1)(s+5a+1)(s+1)}. \tag*{(379)} v s v s + 1 = ( s + 3 a + 1 ) ( s + 5 a + 1 ) ( s + 1 ) ( s + 4 a + 1 ) ( s + a + 1 ) ( s − a + 1 ) . (379)
These ratios imply the identity25
u s = ( 3 a + 1 ) ( 5 a + 1 ) 8 a 3 ( 2 a + 1 ) [ − ( 4 a − 1 ) v s + ( s + 4 a ) v s − ( s + 4 a + 1 ) v s + 1 ] , (380) u_s=\frac{(3a+1)(5a+1)}{8a^3(2a+1)}\left[-(4a-1)v_s+(s+4a)v_s-(s+4a+1)v_{s+1}\right], \tag*{(380)} u s = 8 a 3 ( 2 a + 1 ) ( 3 a + 1 ) ( 5 a + 1 ) [ − ( 4 a − 1 ) v s + ( s + 4 a ) v s − ( s + 4 a + 1 ) v s + 1 ] , (380)
where the last two terms telescope. Summing the above from s = 0 s=0 s = 0 to L L L then gives
∑ s = 0 L u s = ( 3 a + 1 ) ( 5 a + 1 ) 8 a 3 ( 2 a + 1 ) [ − ( 4 a − 1 ) ∑ s = 0 L v s + 4 a − ( L + 1 + 4 a ) v L + 1 ] (381) \sum_{s=0}^{L}u_s=\frac{(3a+1)(5a+1)}{8a^3(2a+1)}\left[-(4a-1)\sum_{s=0}^{L}v_s+4a-(L+1+4a)v_{L+1}\right] \tag*{(381)} s = 0 ∑ L u s = 8 a 3 ( 2 a + 1 ) ( 3 a + 1 ) ( 5 a + 1 ) [ − ( 4 a − 1 ) s = 0 ∑ L v s + 4 a − ( L + 1 + 4 a ) v L + 1 ] (381)
Taking L → ∞ L\to\infty L → ∞ yields
F 5 = ( 3 a + 1 ) ( 5 a + 1 ) 8 a 3 ( 2 a + 1 ) [ 4 a − ( 4 a − 1 ) D ( 4 a + 1 , 1 + a , 1 − a ) ] = ( 3 a + 1 ) ( 5 a + 1 ) 8 a 3 ( 2 a + 1 ) [ 4 a − ( 4 a − 1 ) ( 3 a + 1 2 ) − a − 1 ( 4 a + 2 ) − a − 1 ( 2 a + 3 2 ) − a − 1 ( 5 a + 1 ) − a − 1 ] , (382) \begin{aligned}
F_5=\frac{(3a+1)(5a+1)}{8a^3(2a+1)}\left[4a-(4a-1)\mathcal{D}(4a+1,1+a,1-a)\right] \\
&=\frac{(3a+1)(5a+1)}{8a^3(2a+1)}\left[4a-(4a-1)\frac{\left(3a+\frac{1}{2}\right)_{-a-1}(4a+2)_{-a-1}}{\left(2a+\frac{3}{2}\right)_{-a-1}(5a+1)_{-a-1}}\right], \tag*{(382)}
\end{aligned} F 5 = 8 a 3 ( 2 a + 1 ) ( 3 a + 1 ) ( 5 a + 1 ) [ 4 a − ( 4 a − 1 ) D ( 4 a + 1 , 1 + a , 1 − a ) ] = 8 a 3 ( 2 a + 1 ) ( 3 a + 1 ) ( 5 a + 1 ) [ 4 a − ( 4 a − 1 ) ( 2 a + 2 3 ) − a − 1 ( 5 a + 1 ) − a − 1 ( 3 a + 2 1 ) − a − 1 ( 4 a + 2 ) − a − 1 ] , (382)
where we have used lim L → ∞ ( L + 1 + 4 a ) v L + 1 = 0 \lim_{L\to\infty}(L+1+4a)v_{L+1}=0 lim L → ∞ ( L + 1 + 4 a ) v L + 1 = 0 .
We use eqs. (J.80) and (J.89) to turn eq. (J.79) into
T 1 ( 2 ) = − Y 1 ( 2 ) ( 2 + d ) a ( 2 + d ) − a + ( 2 i N + ) d 4 π a 3 N + ( ( 1 − a 2 ) Γ ( 4 a + 1 ) 2 Γ ( 5 a + 2 ) Γ ( 3 a + 2 ) F 4 + 4 ( 2 a + 1 ) Γ ( 4 a + 1 ) Γ ( 5 a + 2 ) Γ ( 3 a + 2 ) F 5 ) . (383) T_1^{(2)}=-\frac{Y_1^{(2)}}{(2+d)_a(2+d)_{-a}}+\left(\frac{2}{iN_+}\right)^d\frac{4\pi a^3}{N_+}\left(\frac{(1-a^2)\Gamma(4a+1)}{2\Gamma(5a+2)\Gamma(3a+2)}F_4+\frac{4(2a+1)\Gamma(4a+1)}{\Gamma(5a+2)\Gamma(3a+2)}F_5\right). \tag*{(383)} T 1 ( 2 ) = − ( 2 + d ) a ( 2 + d ) − a Y 1 ( 2 ) + ( i N + 2 ) d N + 4 π a 3 ( 2Γ ( 5 a + 2 ) Γ ( 3 a + 2 ) ( 1 − a 2 ) Γ ( 4 a + 1 ) F 4 + Γ ( 5 a + 2 ) Γ ( 3 a + 2 ) 4 ( 2 a + 1 ) Γ ( 4 a + 1 ) F 5 ) . (383)
Then, using the second line of eq. (J.75) for Y 1 ( 2 ) Y_1^{(2)} Y 1 ( 2 ) and eqs. (J.92) and (J.100) for the closed form expressions for F 4 F_4 F 4 and F 5 F_5 F 5 , we reduce and express T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) in a closed form:
T 1 ( 2 ) = − i ( 1 i N + ) 4 a π 3 / 2 Γ ( 2 a − 1 2 ) Γ ( 2 a ) Γ ( a + 1 2 ) Γ ( 3 a − 1 2 ) . (384) T_1^{(2)}=-i\left(\frac{1}{iN_+}\right)^{4a}\pi^{3/2}\frac{\Gamma\left(2a-\frac{1}{2}\right)}{\Gamma(2a)\Gamma\left(a+\frac{1}{2}\right)\Gamma\left(3a-\frac{1}{2}\right)}. \tag*{(384)} T 1 ( 2 ) = − i ( i N + 1 ) 4 a π 3/2 Γ ( 2 a ) Γ ( a + 2 1 ) Γ ( 3 a − 2 1 ) Γ ( 2 a − 2 1 ) . (384)
Quadratic coefficient of β k \beta_k β k
In this appendix, we compute the closed form expression of c 1 c_1 c 1 , the quadratic coefficient of β k \beta_k β k defined in eq. (155) , using eq. (162) with the closed form expressions (J.31) and (J.102) for T 1 ( 0 ) T_1^{(0)} T 1 ( 0 ) and T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) .
We start with expanding eq. (162) to O ( k 2 ) O(k^2) O ( k 2 ) using the splitting (141) for T 1 ( k 2 ) T_1(k^2) T 1 ( k 2 ) :
β k = ( − 1 ) N i N [ ∏ j = 1 N ( ( T 1 ( 0 ) + γ − 2 N ( N + 1 + j ) T 1 ( 2 ) k 2 1 1 0 ) ) ] 21 + O ( k 4 ) (385) \beta_k=(-1)^N i^N\left[\prod_{j=1}^{N}\left(\begin{pmatrix}T_1^{(0)}+\gamma^{-2N(N+1+j)}T_1^{(2)}k^2 & 1 \\ 1 & 0\end{pmatrix}\right)\right]_{21}+O(k^4) \tag*{(385)} β k = ( − 1 ) N i N [ j = 1 ∏ N ( ( T 1 ( 0 ) + γ − 2 N ( N + 1 + j ) T 1 ( 2 ) k 2 1 1 0 ) ) ] 21 + O ( k 4 ) (385)
where we have used eq. (163) . Given eq. (J.5) we note γ − 2 N ( N + 1 + j ) = e 4 i j φ \gamma^{-2N(N+1+j)}=e^{4ij\varphi} γ − 2 N ( N + 1 + j ) = e 4 ij φ , then we rewrite the matrix term in eq. (J.103) as
X j ( k 2 ) ≡ X 0 + e 4 i j φ T 1 ( 2 ) E k 2 , E = ( 1 0 0 0 ) (386) X_j(k^2) \equiv X_0+e^{4ij\varphi}T_1^{(2)}\mathcal{E}k^2,\qquad\mathcal{E}=\begin{pmatrix}1&0\\0&0\end{pmatrix} \tag*{(386)} X j ( k 2 ) ≡ X 0 + e 4 ij φ T 1 ( 2 ) E k 2 , E = ( 1 0 0 0 ) (386)
where X 0 X_0 X 0 is defined in eq. (J.33). Expanding eq. (J.103) with eq. (J.104) then yields c 1 c_1 c 1 :
c 1 = ( − 1 ) N i N [ ∏ j = 1 N ( X 0 + e 4 i j φ T 1 ( 2 ) E k 2 ) ] 21 , k 2 c o e f f = ( − 1 ) N i N T 1 ( 2 ) ∑ j = 1 N e 4 i j φ [ X 0 N − j ] 21 [ E ] 11 [ X 0 j − 1 ] 11 . (387) \begin{aligned}
c_1=(-1)^N i^N\left[\prod_{j=1}^{N}\left(X_0+e^{4ij\varphi}T_1^{(2)}\mathcal{E}k^2\right)\right]_{21,k^2\ \mathrm{coeff}} \\
&=(-1)^N i^N T_1^{(2)}\sum_{j=1}^{N}e^{4ij\varphi}[X_0^{N-j}]_{21}[\mathcal{E}]_{11}[X_0^{j-1}]_{11}. \tag*{(387)}
\end{aligned} c 1 = ( − 1 ) N i N [ j = 1 ∏ N ( X 0 + e 4 ij φ T 1 ( 2 ) E k 2 ) ] 21 , k 2 coeff = ( − 1 ) N i N T 1 ( 2 ) j = 1 ∑ N e 4 ij φ [ X 0 N − j ] 21 [ E ] 11 [ X 0 j − 1 ] 11 . (387)
We use eqs. (J.34) and (J.40) to get
[ X 0 r ] 21 = [ X 0 r − 1 ] 11 = b r = i r − 1 sin ( r φ ) sin φ , r ≥ 1. (388) [X_0^r]_{21}=[X_0^{r-1}]_{11}=b_r=i^{r-1}\frac{\sin(r\varphi)}{\sin\varphi},\qquad r\geq1. \tag*{(388)} [ X 0 r ] 21 = [ X 0 r − 1 ] 11 = b r = i r − 1 sin φ sin ( r φ ) , r ≥ 1. (388)
This turns eq. (J.105) into
c 1 = − T 1 ( 2 ) sin 2 φ Q N , Q N ≡ ∑ j = 1 N e 4 i j φ sin [ ( N − j ) φ ] sin ( j φ ) . (389) c_1=-\frac{T_1^{(2)}}{\sin^2\varphi}Q_N,\qquad Q_N\equiv\sum_{j=1}^{N}e^{4ij\varphi}\sin[(N-j)\varphi]\sin(j\varphi). \tag*{(389)} c 1 = − sin 2 φ T 1 ( 2 ) Q N , Q N ≡ j = 1 ∑ N e 4 ij φ sin [( N − j ) φ ] sin ( j φ ) . (389)
To reduce Q N Q_N Q N , we note an identity N φ = π / 2 − φ N\varphi=\pi/2-\varphi N φ = π /2 − φ from the definition (J.5) of φ \varphi φ , then
sin [ ( N − j ) φ ] sin ( j φ ) = 1 2 [ sin [ ( 2 j + 1 ) φ ] − sin φ ] . (390) \sin[(N-j)\varphi]\sin(j\varphi)=\frac{1}{2}[\sin[(2j+1)\varphi]-\sin\varphi]. \tag*{(390)} sin [( N − j ) φ ] sin ( j φ ) = 2 1 [ sin [( 2 j + 1 ) φ ] − sin φ ] . (390)
Plugging this into Q N Q_N Q N and using e ± i φ = cos φ ± i sin φ e^{\pm i\varphi}=\cos\varphi\pm i\sin\varphi e ± i φ = cos φ ± i sin φ gives
Q N = e i φ 4 i ∑ j = 1 N e 6 i j φ − e − i φ 4 i ∑ j = 1 N e 2 i j φ − sin φ 2 ∑ j = 1 N e 4 i j φ = − e − 2 i φ cos ( 2 φ ) 2 sin ( 3 φ ) . (391) Q_N=\frac{e^{i\varphi}}{4i}\sum_{j=1}^{N}e^{6ij\varphi}-\frac{e^{-i\varphi}}{4i}\sum_{j=1}^{N}e^{2ij\varphi}-\frac{\sin\varphi}{2}\sum_{j=1}^{N}e^{4ij\varphi}=-e^{-2i\varphi}\frac{\cos(2\varphi)}{2\sin(3\varphi)}. \tag*{(391)} Q N = 4 i e i φ j = 1 ∑ N e 6 ij φ − 4 i e − i φ j = 1 ∑ N e 2 ij φ − 2 sin φ j = 1 ∑ N e 4 ij φ = − e − 2 i φ 2 sin ( 3 φ ) cos ( 2 φ ) . (391)
Hence, plugging Q N Q_N Q N and the closed form expression (J.102) for T 1 ( 2 ) T_1^{(2)} T 1 ( 2 ) into c 1 c_1 c 1 given by eq. (J.107), we get the closed form expression
c 1 = i π 3 / 2 N + 2 ( 1 − 1 N + ) Γ ( 2 a − 1 2 ) Γ ( 2 a ) Γ ( a + 1 2 ) Γ ( 3 a − 1 2 ) cos ( 2 φ ) 2 sin 2 ( φ ) sin ( 3 φ ) . (392) c_1=i\frac{\pi^{3/2}}{N_+^{2(1-\frac{1}{N_+})}}\frac{\Gamma(2a-\frac{1}{2})}{\Gamma(2a)\Gamma(a+\frac{1}{2})\Gamma(3a-\frac{1}{2})}\frac{\cos(2\varphi)}{2\sin^2(\varphi)\sin(3\varphi)}. \tag*{(392)} c 1 = i N + 2 ( 1 − N + 1 ) π 3/2 Γ ( 2 a ) Γ ( a + 2 1 ) Γ ( 3 a − 2 1 ) Γ ( 2 a − 2 1 ) 2 sin 2 ( φ ) sin ( 3 φ ) cos ( 2 φ ) . (392)
where we have used φ = π / ( 2 N + ) \varphi=\pi/(2N_+) φ = π / ( 2 N + ) and a = N / ( 2 N + ) a=N/(2N_+) a = N / ( 2 N + ) to compute the phase factor. Here we remark that c 1 c_1 c 1 is purely imaginary, explicitly confirming the time-reversal symmetry result (85) . Further simplifying eq. (J.110) using the gamma function identity Γ ( z ) Γ ( 1 − z ) = π / sin ( π z ) \Gamma(z)\Gamma(1-z)=\pi/\sin(\pi z) Γ ( z ) Γ ( 1 − z ) = π / sin ( π z ) and restoring the index notations a = N / ( 2 ( N + 1 ) ) a=N/(2(N+1)) a = N / ( 2 ( N + 1 )) , φ = π / ( 2 ( N + 1 ) ) \varphi=\pi/(2(N+1)) φ = π / ( 2 ( N + 1 )) and N + = N + 1 N_+=N+1 N + = N + 1 yields the closed form expression
c 1 ( n ) = i cos [ π 2 ( N + 1 ) ] Γ [ 1 2 ( N + 1 ) ] Γ ( 1 N + 1 ) Γ [ 3 2 ( N + 1 ) ] π ( N + 1 ) 2 ( 1 − 1 N + 1 ) Γ ( 1 N + 1 + 1 2 ) . (393) c_1(n)=i\frac{\cos\left[\frac{\pi}{2(N+1)}\right]\Gamma\left[\frac{1}{2(N+1)}\right]\Gamma\left(\frac{1}{N+1}\right)\Gamma\left[\frac{3}{2(N+1)}\right]}{\sqrt{\pi}(N+1)^{2(1-\frac{1}{N+1})}\Gamma\left(\frac{1}{N+1}+\frac{1}{2}\right)}. \tag*{(393)} c 1 ( n ) = i π ( N + 1 ) 2 ( 1 − N + 1 1 ) Γ ( N + 1 1 + 2 1 ) cos [ 2 ( N + 1 ) π ] Γ [ 2 ( N + 1 ) 1 ] Γ ( N + 1 1 ) Γ [ 2 ( N + 1 ) 3 ] . (393)
This is eq. (J.1), the main result of this appendix.