WorksPhysics and Astronomy
Exact thermal correlators at vanishing momenta and the black hole singularity
Abstract
We study thermal two- and three-point functions of scalar primaries in holographic CFTs dual to AdS black branes. The three-point function arises from a cubic scalar interaction in the bulk. At vanishing external momenta, the tree-level results can be written in closed form and are exact in the operator dimensions. We reproduce them through direct CFT calculations in two dimensions. Their large-dimension expansions contain subleading exponential terms encoding the radial proper time from the horizon to the black hole singularity. These results provide an exact test of earlier WKB and saddle-point predictions for higher-point thermal correlators, confirming the proper-time signature while correcting the predicted asymptotic form.
Introduction
The AdS/CFT correspondence [1, 2, 3] relates thermal CFT correlators to bulk observables in black hole geometries. The Euclidean setup contains only the exterior region, raising the question of how CFT observables encode the interior and its singularity. This paper tests a specific prediction for higher-point thermal correlators: whether their large-dimension behaviour contains the proper time from the horizon to the black hole singularity.
In [4], I studied this question for scalar self-interactions and exchange interactions using WKB and saddle-point approximations. The predicted leading exponential combined the geodesic distance between the boundary and the horizon with the proper time to the singularity. Here, we test the contact three-point prediction at vanishing external momenta in the black brane geometry, where the bulk propagator and vertex integral can be treated exactly in the operator dimension. The resulting large-dimension expansion encodes the time to singularity in subleading exponential terms, while the leading exponential depends on the boundary-to-horizon distance. This provides a direct benchmark for the earlier approximation in the zero-momentum case.
Our main results are the exact identical and mixed scalar contact three-point functions in general boundary dimension , together with their analytically continued large-dimension limits. We also rederive the thermal two-point function obtained by Berenstein and Mancilla [5] and analyze its asymptotic behaviour. We independently derive the results by direct CFT calculation of the zero-momentum correlators.
The proper-time interpretation is motivated by exact one-point calculations. Grinberg and Maldacena [6] computed the thermal one-point function induced by coupling a bulk scalar to the square of the Weyl tensor. Its analytically continued large-mass limit contains the proper time from the horizon to the singularity and agrees with a WKB and saddle-point calculation. David and Kumar [7] obtained further exact results for charged black branes in a large bulk spacetime dimension limit and for a hyperbolic black hole, with appropriate curvature couplings.1 Related signatures occur in curvature-induced thermal two-point contributions [8, 9]. These examples motivate testing the higher-point approximation against the exact contact contributions considered here.
A complementary approach probes singularities by analytically continuing boundary correlators [10, 11, 12]. For AdS–Schwarzschild black holes in four and higher dimensions, the geodesic approximation reveals a singularity on a secondary sheet of the two-sided correlator, associated with “bouncing geodesics”: nearly null spacelike geodesics that connect the two boundaries and approach the black hole singularity. Related signatures have been studied through the thermal OPE [13], spectral functions [14], large-frequency two-point functions [15], line defects [16], and quasinormal modes [17, 18]. Extensions to shock-wave geometries [19] and black holes with inner horizons [20, 21] have also been studied. For a charged black brane, exponentially small terms in the spectral density at large real frequency have been related to null geodesics reflecting off its timelike singularity [22]. Analytically continued, smeared thermal four-point functions provide a related probe of local scattering behind the horizon [23]. The fate of bouncing singularities has also been examined at finite coupling in SYK models [24, 25], with finite string length effects in the NS5-brane worldsheet coset model [26], and at finite [27, 28, 29].
The remainder of the paper is organized as follows. § 2 fixes the conventions and collects the results. The bulk calculation and large-dimension analysis are performed in § 3. § 5 gives the CFT check in two dimensions. We conclude with a short discussion of the comparison with the previous approximation and open questions in § 6.
Preliminaries and results
This section describes the setup, fixes notation, and collects the results obtained in this paper. The physical observables of interest in this work are thermal correlation functions of primary operators in holographic CFTs. Specifically, we consider a Euclidean CFT on , where is the inverse temperature. The bulk dual is a black brane geometry with Hawking temperature . We use for a scalar primary operator of conformal dimension . Its dual is a scalar field propagating in the bulk spacetime, with mass related to the conformal dimension by [2, 3]
where is the AdS radius.
The translation symmetries of the background make it convenient to work with correlators in Fourier space. The -point thermal correlator of is denoted by2
Here, are the momenta, including the Matsubara frequencies. Translation symmetry enforces momentum conservation in the correlator (2). Therefore, following the convention of [30], we define a reduced correlator via the relation
where it is understood that time translation symmetry on requires the sum over Matsubara frequencies to vanish:
We set all external momenta to zero and use the exact bulk-boundary propagator in this limit. The thermal two-point function is given in (19); the identical and mixed contact three-point functions (see Figure 1 and Figure 2) are given in (30) and (40). § 5 checks these expressions using a direct CFT computation in two dimensions.
Figure 1. The contact Witten diagram sourcing the three-point function of a scalar primary . The blue edges correspond to bulk-boundary propagators of the dual field . The interaction vertex is integrated over the bulk Euclidean geometry.
Figure 2. The contact Witten diagram sourcing the mixed three-point function of two primaries and . The edges correspond to bulk-boundary propagators of the dual fields: blue for and red for . The interaction vertex is integrated over the bulk Euclidean geometry.
The correlators have poles for real positive values of . So, to analyze the large conformal dimension behaviour, we give a negative imaginary part and take the limit . The resulting expressions, given in (22) and (32), are:
where , with being the time to singularity and the regularized distance between the asymptotic boundary and the black hole horizon; see equations (23) and (24). In this sense, the thermal correlators at vanishing momenta probe the bulk geometry, including the black hole interior and singularity. The mixed thermal three-point function of two different primaries shows a similar signature when one of the operator dimensions is taken large; see (43).
Gravity calculation
The bulk geometry dual to a holographic CFT on is a Euclidean black brane whose metric is given by
Here, is the AdS radius and is the location of the horizon. The asymptotic boundary is at . We consider the following action for a bulk scalar field , dual to a boundary primary operator :
The relation between the mass of the bulk scalar and the conformal dimension of the boundary operator was given in (1). The self-interaction in (8) provides a cubic vertex for Witten diagrams [2, 3].
Let us first focus on the free (quadratic) part of the action, which gives the Klein–Gordon equation for the massive scalar field. In momentum space, with , it reduces to a radial equation:
The bulk-boundary propagator, , is defined as a regular solution to this equation with the boundary condition
At vanishing momenta, the propagator is known in closed form as a hypergeometric function [6]:
where and we have defined a new radial variable
An alternative form of the propagator uses the Legendre function [31]:
Two-point function
Following the standard rules of AdS/CFT [32, 33] and holographic renormalization [34], the thermal two-point function of the operator can directly be extracted from the bulk-boundary propagator. This computation appeared in [5]. The idea is to expand the propagator around as
where the denote higher powers of . is termed the normalizable coefficient. The renormalized two-point function is then given by
Using the connection formula for the hypergeometric function [35 equation (15.8.4)], we can extract the near-boundary behaviour of the bulk-boundary propagator at zero momentum:
Substituting this in (11) and reading off the coefficient of gives the normalizable coefficient :
We then have the two-point function at vanishing momenta
This reproduces the static response function obtained in equations (25)–(26) of [5]. Using the reflection formula, , we write (18) as
Large conformal dimension limit
The thermal correlator (19), as a function of the conformal dimension, has poles at
To avoid these poles in the asymptotic limit, we give a small negative imaginary part to (equivalently ). Since the spacetime dimension is fixed, this is implemented by . It is convenient to define the following variable to organize the expansion:
Keeping terms exponential in and suppressing the temperature dependence, the asymptotic form of the two-point function is
We can identify the factors appearing in the expansion with geometric quantities in the black brane background. The regularized distance between the asymptotic boundary and the horizon is
The proper time between the horizon and black hole singularity along a radially infalling geodesic is
Thus, at large mass, the expansion (22) can be expressed as
where it is understood that the coefficient of each exponential factor inside the brackets is a power series in .
Three-point function
The cubic self-interaction vertex in (8) sources the three-point function of the operator . The contact Witten diagram, shown in Figure 1, evaluates to
Substituting (13) for the zero-momentum propagator and changing the integration variable, we obtain the correlator at vanishing momenta:
The near-boundary behaviour of the integrand is as . This means that the integral converges only if . For larger values of , the integral is defined by analytic continuation in .
Reference [36] computed the integral of three associated Legendre functions of the same order.3 We adapt the general formula given in equations (1)–(2) of [36] to the integral of three Legendre functions:
where each in the arguments on the right-hand side denotes a product of Gamma functions over the different choices. So, the numerator of (28) contains eight Gamma functions while the denominator contains six. The integral in (27) is obtained by setting . Using the reflection identity and the duplication formula, , we can simplify the integral as follows:
Substituting (29) in (27) gives the thermal three-point function:
Large conformal dimension limit
As with the two-point function, we now study the large conformal dimension behaviour of the thermal three-point function (30). From the ratio of sine functions, we identify the poles of the correlator at positive real values of the conformal dimension:4
where is an integer not divisible by 3. Again, to avoid these poles, we analyze the limit of the correlator. The terms exponential in the conformal dimension come from the ratios of Gamma functions and sine functions. In the variable defined in (21), the thermal three-point function is expanded as
The expansion can be re-expressed using the geometric quantities (23) and (24) as
This indicates that higher-point thermal correlators contain an imprint of the black hole singularity. However, the result (33) differs from the form suggested in [4]. In fact, the expansion (33) has no terms corresponding to for any positive integer . See § 6 for further discussion.
Mixed correlator
In [4], it was argued that factors of time to singularity may also appear in mixed thermal correlators of heavy and light operators. We can test this by directly calculating the following three-point function at vanishing momenta:
where and are two scalar primaries holographically dual to scalar fields and , respectively. The conformal dimensions, , are related to the masses, , of the dual scalars by the usual relation (1). We consider the bulk action
The interaction provides a cubic vertex for Witten diagrams. The contact diagram using this vertex sources the three-point function , as shown in Figure 2. It evaluates to the bulk integral
where we have denoted the bulk-boundary propagator of by . Using the zero-momentum propagators (13) and changing the integration variable, we express the correlator (34) as
Near-boundary behaviour implies that the integral converges for . For other values, we define it by analytic continuation in and .
The general formula (28) can be used for (37) by setting and . The resulting ratio of Gamma functions can be simplified as follows:
where we have defined
Substituting (38) in (37) gives the thermal mixed correlator:
where we have introduced
Large conformal dimension limit
We wish to analyze the behaviour of the mixed correlator at large while holding fixed. For generic values of , the correlator has poles at
where is an integer such that the BF bound, , is satisfied. As before, we avoid the poles by taking the limit along a ray in the lower half of the complex plane, . The expansion of the mixed correlator begins as
where we have defined
Identifying the geometric quantities (23) and (24), we write
We see that the thermal mixed correlator also contains a signature of the black hole singularity. The precise form (45) differs from the one suggested in [4], where the expected form was .
CFT
In , thermal correlators on follow from the vacuum ones on the plane by a conformal map; see equation (46). The functional form of the two- and three-point correlators on the plane can be fixed completely using conformal symmetry. Moreover, the remaining data — normalization and three-point OPE coefficients — can be computed for a holographic theory using bulk Witten diagrams in the vacuum AdS geometry. Momentum-space thermal three-point functions in two dimensions were previously computed in [38] with Lorentzian ordering prescriptions. Here, we evaluate the zero-momentum Euclidean correlators, providing a separate check of the bulk formulas.
Let , be coordinates on the cylinder , with and . The conformal map
takes the cylinder to the complex plane .5 A primary with conformal weights transforms under the map (46) as:
We now restrict to primaries with equal holomorphic and antiholomorphic weights: , where, for a holographic theory, is given by (1). Further, we will use the compact notation and for operators on the plane and cylinder, respectively.
Two-point function
We start with the vacuum two-point function of on the plane, which, by conformal invariance, is given by
where . For a holographic theory, the normalization of the two-point function can be computed using the bulk-boundary propagator in vacuum AdS [32 equation (96)]:
Applying the conformal transformation (47) to (48) gives the thermal two-point function in position space:
While this expression can be further simplified to [33]
with , we will find it more convenient to work with (50).
The translation symmetry on the cylinder allows us to define the reduced Fourier space thermal two-point function
where . Recall that we are interested in thermal correlators at vanishing momenta. Substituting (50) into (52) and using the change of variables (46) from to gives
where . The integral in (53) is an instance of the complex Selberg integral [39], also sometimes referred to as the Dotsenko–Fateev integral [40]:
The general formula for can be written as [41, 42]
The defining integral (54) is convergent in a “narrow domain” [41, 42], but the result (55) in terms of the Gamma and trigonometric functions can be analytically continued to all complex parameter values.
The complex Selberg integral relevant for the thermal two-point function (53) is .6 Using the general formula (55), its value is given by
where we have used the reflection formula to simplify the expression. Substituting the formula back into (53) and using the holographic value (49) for the normalization gives
This precisely matches the gravity result (19) for , in which case the black brane radius is related to the inverse temperature by .
Three-point function
We now turn to the three-point function of the primary . Again, by conformal invariance, the vacuum three-point correlator on the plane takes the form
where . The factor is related to the OPE coefficient of the operator . For a holographic theory, it can be computed in the semiclassical limit by bulk perturbation theory in the vacuum AdS geometry. The vacuum contact Witten diagram (with the same schematic form as Figure 1) corresponding to a cubic interaction of bulk scalars was computed in [32]. Specializing their formula [32 equation (25)] to the self-interacting scalar in gives
We apply the conformal transformation (47) to (58) to obtain the thermal three-point function in position space:
This formula can be expressed as
with , but it is easier to work with (60) when taking the Fourier transform.
The reduced Fourier space thermal three-point function is defined as
For vanishing momenta, we get
The above integral is the complex Selberg integral . Specializing the general formula (55) to this case and using the reflection identity for the Gamma function, we obtain
Substituting the formula back into (63) and using the holographic value (59) for gives the thermal three-point function
This is in precise agreement with the gravity result (30) for and .
Mixed correlator
We now repeat the CFT calculation for the mixed three-point function (34). Conformal invariance fixes the vacuum form to
where the vacuum contact diagram for the interaction in (35) gives [32]
Applying (47) gives the thermal correlator. Its zero-momentum Fourier transform, as before, is given by a complex Selberg integral:
Using (55) and the reflection formula gives
with defined in (39). Combining the result with (67), we find
where is given in (41). This precisely reproduces (40) at and .
Discussion
We have computed thermal two- and three-point correlators at vanishing external momenta in the black brane background. The two-point function follows from the normalizable coefficient of the exact bulk-boundary propagator, while the contact three-point diagrams reduce to integrals of three Legendre functions. These integrals give closed-form expressions both for identical scalar primaries and for the mixed correlator with two identical operators and a third of arbitrary dimension. The results are exact in the operator dimensions and temperature at the bulk orders considered: the free two-point function and the tree-level three-point contributions. In , we independently reproduced all three expressions by mapping vacuum CFT correlators to the thermal cylinder and evaluating their zero-momentum Fourier transforms as complex Selberg integrals.
We identified the poles arising from the trigonometric dependence of the zero-momentum thermal correlators on the operator dimension. These poles are isolated on the positive real axis. Analogous poles were seen in the thermal one-point function in the black brane background [5, 6].7 Those poles were interpreted, in the semiclassical large- limit, as a consequence of enhanced operator mixing at resonant values of the operator dimension. It would be interesting to develop a similar interpretation of the poles in higher-point thermal correlators.
By analytically continuing the conformal dimension to the lower half of the complex plane, we analyzed the large conformal dimension behaviour of the thermal correlators. We showed that their asymptotic forms encode geometric information about the black hole, including a signature of its singularity. More precisely, the leading exponential mass dependence is for the two-point function and for the three-point function, where is the regularized geodesic distance between the horizon and the asymptotic boundary. The proper time to singularity, , appears through powers of in the subleading exponential terms.
The exact three-point result for identical primaries disagrees with the leading form predicted in [4 equation (4.7)] for the continued zero-momentum contact contribution. The earlier calculation retained the WKB prefactors and included a family of complex saddles near the singularity, but left the horizon endpoint contribution uncontrolled because the WKB approximation breaks down there. The discrepancy may arise from this endpoint contribution as well as from an incomplete account of the relevant saddles and their contributions. The present calculation does not show how to repair the failure of the WKB calculation in the higher-point function, nor does it explain why the one-point saddle calculation succeeds. However, the asymptotic expansion constrains how the proper time to the singularity enters: the leading exponential depends only on the boundary-to-horizon length, and terms of the form are absent for every positive integer . An exact WKB analysis [43, 44] is a natural next step to address these questions.8
Acknowledgments
I am grateful to Justin R. David for important comments and prompt feedback on the original calculations. I thank R. Loganayagam for several discussions on this work. I also thank Saikat Ghosh, Ashoke Sen, Spenta Wadia, and the rest of the ICTS string group for various interactions during the course of this work. I acknowledge support from the Department of Atomic Energy, Government of India, under project no. RTI4019.
Statement on AI use. I used ChatGPT for some editing, figures, typesetting, and verification of certain claims in this work. The manuscript was written by me and I take full responsibility for any errors.
References
- [1]J.M. Maldacena, The Large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2 (1998) 231 [hep-th/9711200].
- [2]S.S. Gubser, I.R. Klebanov and A.M. Polyakov, Gauge theory correlators from noncritical string theory, Phys. Lett. B 428 (1998) 105 [hep-th/9802109].DOI
- [3]E. Witten, Anti de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253 [hep-th/9802150].arxiv.org/abs/hep-th/9802150
- [4]K. Singhi, Proper time to the singularity and thermal correlators, Phys. Rev. D 112 (2025) 106011 [2406.08553].DOI
- [5]D. Berenstein and R. Mancilla, Aspects of thermal one-point functions and response functions in AdS black holes, Phys. Rev. D 107 (2023) 126010 [2211.05144].DOI
- [6]M. Grinberg and J. Maldacena, Proper time to the black hole singularity from thermal one-point functions, JHEP 03 (2021) 131 [2011.01004].
- [7]J.R. David and S. Kumar, Thermal one point functions, large d and interior geometry of black holes, JHEP 03 (2023) 256 [2212.07758].DOI
- [8]D. Rodriguez-Gomez and J.G. Russo, Correlation functions in finite temperature CFT and black hole singularities, JHEP 06 (2021) 048 [2102.11891].
- [9]H. Krishna and D. Rodriguez-Gomez, Holographic thermal correlators revisited, JHEP 11 (2021) 139 [2108.00277].DOI
- [10]P. Kraus, H. Ooguri and S. Shenker, Inside the horizon with AdS / CFT, Phys. Rev. D 67 (2003) 124022 [hep-th/0212277].DOI
- [11]L. Fidkowski, V. Hubeny, M. Kleban and S. Shenker, The Black hole singularity in AdS / CFT, JHEP 02 (2004) 014 [hep-th/0306170].
- [12]G. Festuccia and H. Liu, Excursions beyond the horizon: Black hole singularities in Yang-Mills theories. I., JHEP 04 (2006) 044 [hep-th/0506202].
- [13]N. Čepļak, H. Liu, A. Parnachev and S. Valach, Black Hole Singularity from OPE, JHEP 10 (2024) 105 [2404.17286].arxiv.org/abs/2404.17286
- [14]H.F. Jia and M. Rangamani, Exact holographic thermal spectral functions: OPE, non-perturbative corrections, and black hole singularity, 2604.10803.
- [15]N. Afkhami-Jeddi, S. Caron-Huot, J. Chakravarty and A. Maloney, Imprint of the black hole singularity on thermal two-point functions, Phys. Rev. D 114 (2026) 025005 [2510.21673].
- [16]S. Giombi, Y.-Z. Li and J. Shan, Bouncing singularities and thermal correlators on line defects, JHEP 07 (2026) 170 [2603.11012].
- [17]G. Festuccia and H. Liu, A Bohr-Sommerfeld quantization formula for quasinormal frequencies of AdS black holes, Adv. Sci. Lett. 2 (2009) 221 [0811.1033].DOI
- [18]P. Arnaudo and B. Withers, Bouncing singularities in Schwarzschild: a geometric origin of the QNM convergence region, 2605.16489.arxiv.org/abs/2605.16489
- [19]G.T. Horowitz, H. Leung, L. Queimada and Y. Zhao, Boundary signature of singularity in the presence of a shock wave, SciPost Phys. 16 (2024) 060 [2310.03076].DOI
- [20]N. Čepļak, H. Liu, A. Parnachev and S. Valach, Fooling the censor: going beyond inner horizons with the OPE, JHEP 05 (2026) 001 [2511.09638].arxiv.org/abs/2511.09638
- [21]S. Ali Ahmad, A. Almheiri and S. Lin, Continuing past the inner horizon using WKB, 2601.02354.arxiv.org/abs/2601.02354
- [22]J. Chakravarty and Y. Zిగdon, A signature of the timelike singularity in holography, 2610.30134.arxiv.org/abs/2610.30134
- [23]J. Chakravarty, Imprint of black hole interior on thermal four-point correlators, JHEP 09 (2026) 034 [2512.10912].arxiv.org/abs/2512.10912
- [24]M. Dodelson, C. Iossa and R. Karlsson, Bouncing off a stringy singularity, JHEP 07 (2026) 161 [2511.09616].
- [25]I. Burić, C.-M. Chang, I. Gusev, E. Helfenberger, A. Parnachev and M. Rangamani, Thermal two-point functions in SYK and complex-time singularities, 2607.05258.arxiv.org/abs/2607.05258
- [26]I. Burić, I. Gusev and A. Parnachev, Stringy resolution of bouncing singularities, 2610.12311.arxiv.org/abs/2610.12311
- [27]Z. Komargodski, Fate of Bouncing Singularities at Finite G_N, 2609.31309.arxiv.org/abs/2609.31309
- [28]G. Batra, A. Levine and S.H. Shenker, A backreacted bouncing geodesic in 2D, 2609.32071.arxiv.org/abs/2609.32071
- [29]M. Liu and S.-M. Ruan, Smoothing the Bouncing-Geodesic Singularity in AdS/CFT, 2611.03347.arxiv.org/abs/2611.03347
- [30]A. Bzowski, P. McFadden and K. Skenderis, Implications of conformal invariance in momentum space, JHEP 03 (2014) 111 [1304.7760].
- [31]A. Yarom, Notes on the bulk viscosity of holographic gauge theory plasmas, JHEP 04 (2010) 024 [0912.2100].DOI
- [32]D.Z. Freedman, S.D. Mathur, A. Matusis and L. Rastelli, Correlation functions in the CFT(d) / AdS(d+1) correspondence, Nucl. Phys. B 546 (1999) 96 [hep-th/9804058].DOI
- [33]D.T. Son and A.O. Starinets, Minkowski space correlators in AdS / CFT correspondence: Recipe and applications, JHEP 09 (2002) 042 [hep-th/0205051].
- [34]K. Skenderis, Lecture notes on holographic renormalization, Class. Quant. Grav. 19 (2002) 5849 [hep-th/0209067].DOI
- [35]“NIST Digital Library of Mathematical Functions.” https://dlmf.nist.gov/, Release 1.2.4 of 2025-03-15.
- [36]J. Bros, H. Epstein, M. Gaudin, U. Moschella and V. Pasquier, Triangular invariants, three-point functions and particle stability on the de Sitter universe, Commun. Math. Phys. 295 (2010) 261 [0901.4223].DOI
- [37]P. Breitenlohner and D.Z. Freedman, Positive energy in anti-de Sitter backgrounds and gauged extended supergravity, Physics Letters B 115 (1982) 197.DOI
- [38]M. Becker, Y. Cabrera and N. Su, Finite-temperature three-point function in 2D CFT, JHEP 09 (2014) 157 [1407.3415].
- [39]K. Aomoto, On the complex Selberg integral, Quart. J. Math. Oxford Ser. (2) 38 (1987) 385.
- [40]V.S. Dotsenko and V.A. Fateev, Four-point correlation functions and the operator algebra in 2D conformal invariant theories with central charge C ≤ 1, Nucl. Phys. B 251 (1985) 691.DOI
- [41]P.J. Forrester and S.O. Warnaar, The importance of the Selberg integral, Bull. Amer. Math. Soc. (N.S.) 45 (2008) 489 [0710.3981].arxiv.org/abs/0710.3981
- [42]Y.A. Neretin, On the Dotsenko–Fateev complex twin of the Selberg integral and its extensions, Ramanujan J. (2023) [2212.09112].DOI
- [43]T. Kawai and Y. Takei, Algebraic Analysis of Singular Perturbation Theory, Iwanami series in modern mathematics, American Mathematical Society (2005).
- [44]K. Iwaki and T. Nakanishi, Exact WKB analysis and cluster algebras, J. Phys. A 47 (2014) 474009 [1401.7094].arxiv.org/abs/1401.7094

