betaFeedback

WorksPhysics and Astronomy

Triple Exchange at the Cosmological Collider in Closed Form

Abstract

Particles with masses near the Hubble scale leave a characteristic pattern in the primordial bispectrum, turning inflation into a cosmological collider. For light fields, where this signal is largest, no closed form is known for the full shape of the quasi-single-field triple exchange. Using Feynman-integral methods, we obtain the first closed-form results, exact in the momenta, as expansions in terms of multiple polylogarithms around the massless and conformally coupled limits. We show that the template used in Planck analyses overestimates the amplitude by up to a factor of 1.8 and underestimates the mass, whereas our new template recovers both, as needed for current and upcoming CMB and galaxy surveys. We also compute the collapsed trispectrum and, combining it with the bispectrum, find that at weak mixing it bounds the self-coupling more tightly.

Introduction

The anisotropies of the cosmic microwave background (CMB) and the large-scale structure of the Universe originate from quantum fluctuations that inflation stretched to cosmological scales. Their statistics probe physics at the Hubble scale HH of inflation, up to 101310^{13} GeV [1], far beyond the reach of colliders. While the power spectrum is measured to percent precision [1], the particle content of inflation is encoded in primordial non-Gaussianity [2, 3], most directly in the bispectrum. In particular, a field of mass m∼Hm \sim H leaves a distinctive imprint in its squeezed limit, a non-analytic power law in the momentum ratio for light fields, m<32Hm < \frac{3}{2}H, and an oscillation in its logarithm for heavier ones, with an exponent or a frequency set by the mass [3, 4, 5, 6] [7]. This cosmological collider signal offers a unique opportunity to search for new particles at the Hubble scale.

The search for this signal has begun. Analyses of Planck data [8, 9, 10, 11, 12] and of galaxy clustering [13, 14] have found no evidence so far, and the Simons Observatory [15] and SPHEREx [16] will tighten the bounds. Since these searches fit templates, fixed functions of the momenta whose amplitude fNLf_{\rm NL} is measured, translating a bound or detection into the mass and couplings of the new field requires the exact shape or a calibrated template.

For the original model of the cosmological collider, quasi-single-field (QSF) inflation [3, 17], this calibration has long been missing where it matters most. In this model, a scalar of mass m∼Hm \sim H mixes quadratically with the inflaton and has a cubic self-interaction, which at weak mixing contributes through the triple exchange of Figure 1. While the single and double exchanges are known exactly [18, 19, 20, 21, 22], the latter also with high precision in the physical region [23], the triple exchange is known numerically [1, 24, 25] (see [26] for general trees), as a one-dimensional integral [27], as finite Mellin-Barnes (MB) integrals for heavy fields [28], and as hypergeometric series [29, 30, 31] that diverge at physical kinematics [30, 31]. Moreover, with the mixing resummed, it is expressed as integrals at any mass [32, 33, 34], as series that converge

Figure 1. The triple-exchange diagram, in which three massive lines (red), coupled to inflaton legs (blue) at mixing vertices (crosses), meet at a cubic vertex (dot).

above the conformally coupled mass [35] and, very recently, as series in the momentum ratios that converge at physical kinematics for any mass and mixing [36]. These representations are exact in the mass. However, for light fields, which decay slowly outside the horizon and give the largest signal [3, 37], the complementary information, the momentum dependence in closed form, has not been available, and the Planck analyses [8, 9] relied on an approximate template [3] never calibrated to infer the coupling and the mass.

In this Letter, we use methods from Feynman integral calculus to obtain, for the first time, the QSF triple exchange of light fields as expansions in the mass that are exact in the momenta, by two independent routes. Their coefficients are expressed in terms of multiple polylogarithms [38] (MPLs) whose letters reveal the analytic structure in the momenta. The two expansions describe one function that is meromorphic in the mass, with poles only at m2≤0m^2 \le0, and extend to a finite duration of inflation. The QSF template, used without calibration, biases the estimated coupling by factors up to 1.8 and underestimates the mass, and we provide a simple template that measures both. Finally, the collapsed trispectrum can separate the mixing from the self-coupling, and we combine the Planck bispectrum and trispectrum [39] constraints (Figure 2).

Figure 2. Planck upper bounds on the self-coupling of a light field versus its mixing, at m=0.38Hm = 0.38H (solid) and 1.02H1.02H (dashed), from the bispectrum (blue) and the collapsed trispectrum [39] (orange, with a band spanning the two component-separation pipelines). Left of each crossing (dot for m=0.38Hm = 0.38H), the trispectrum gives the stronger bound.

**Setup—**In the decoupling limit of the effective field theory of inflation [40], the canonically normalized Goldstone boson πc\pi_c of broken time translations couples to a scalar σ\sigma of mass mm through the QSF interactions [3]

La3⊃ρπ˙cσ−μσ3,(1)\frac{\mathcal{L}}{a^3} \supset\rho\dot{\pi}_c \sigma-\mu\sigma^3 , \tag*{(1)}

with aa the scale factor, dots denoting derivatives with respect to cosmic time, and ρ\rho and μ\mu the mixing and cubic couplings. In the conventions of [41], the curvature perturbation is ζ=−πc/(2ϵMpl)\zeta=-\pi_c/(\sqrt{2\epsilon}M_{\mathrm{pl}}), with ϵ\epsilon the first slow-roll parameter and MplM_{\mathrm{pl}} the reduced Planck mass. At order ρ3μ\rho^3\mu, (1) gives the diagram of Figure 1, and we compute the late-time bispectrum BπB_{\pi} in the Bunch–Davies vacuum [42], defined by ⟨πc,k1πc,k2πc,k3⟩=(2π)3δ3(k1+k2+k3)Bπ\langle\pi_{c,\mathbf{k}_1}\pi_{c,\mathbf{k}_2}\pi_{c,\mathbf{k}_3}\rangle=(2\pi)^3\delta^3(\mathbf{k}_1+\mathbf{k}_2+\mathbf{k}_3)B_{\pi}, with ki=∣ki∣k_i=\lvert\mathbf{k}_i\rvert. Setting H=1H=1 and writing m2=94−ν2m^2=\frac{9}{4}-\nu^2, we define the shape function [43]

Sπ≡(k1k2k3)2Bπ=−34ρ3μk1k2k3A3,(2)S_{\pi}\equiv(k_1k_2k_3)^2B_{\pi}=-\frac{3}{4}\rho^3\mu k_1k_2k_3A_3 , \tag*{(2)}

where A3(k1,k2,k3;ν)A_3(k_1,k_2,k_3;\nu) is homogeneous of degree −3-3 in the momenta, and the constant −34-\frac{3}{4} normalizes its massless limit [see Eq. (9)].

In the in-in formalism [44], the four vertices are integrated over both branches of the Schwinger–Keldysh contour. After Wick rotation of the conformal time, η→±it\eta\to\pm it with t>0t>0 on the two branches [45], two kinds of σ\sigma lines appear [28]. An uncut line of momentum kk between the times tt and t′t' carries the time-ordered kernel Kν(kt>)Iν(kt<)K_{\nu}(kt_>)I_{\nu}(kt_<) of modified Bessel functions, whereas a cut line factorizes into Kν(kt)Kν(kt′)K_{\nu}(kt)K_{\nu}(kt'). Including the measure and the external legs, a mixing vertex at time tlt_l is integrated with dtl tl−1/2e−kltldt_l\,t_l^{-1/2}e^{-k_lt_l}, and the cubic vertex at time t0t_0 with dt0 t01/2dt_0\,t_0^{1/2} [A]. The line of momentum klk_l therefore enters through one of two functions of t0t_0,

Ll(t0)=∫0∞dtltle−kltlKν(klt>)Iν(klt<),L_l(t_0)=\int_0^\infty\frac{dt_l}{\sqrt{t_l}}e^{-k_lt_l}K_{\nu}(k_lt_>)I_{\nu}(k_lt_<),
Cl(t0)=Kν(klt0),(3)C_l(t_0)=K_{\nu}(k_lt_0), \tag*{(3)}

for an uncut and a cut line, with t>t_> and t<t_< the larger and the smaller of t0t_0 and tlt_l. These define the four blocks of the cutting rule of [28] (see also [46]),

ψ3=⟨L1L2L3⟩,ψ2[l]=⟨ClLmLn⟩,\psi_3=\langle L_1L_2L_3\rangle,\qquad\psi_2[l]=\langle C_lL_mL_n\rangle,
ψ1[l]=⟨LlCmCn⟩,ψ0=⟨C1C2C3⟩,(4)\psi_1[l]=\langle L_lC_mC_n\rangle,\qquad\psi_0=\langle C_1C_2C_3\rangle, \tag*{(4)}

where ⟨f⟩≡∫0∞dt0 t01/2f(t0)\langle f\rangle\equiv\int_0^\infty dt_0\,t_0^{1/2}f(t_0), {l,m,n}\{l,m,n\} is a permutation of {1,2,3}\{1,2,3\}, and ψn\psi_n has nn uncut lines. Outside their convergence domain, we define these integrals by analytic continuation in ν\nu. Cutting the line ll replaces Ll(t0)L_l(t_0) by ΦlCl(t0)\Phi_lC_l(t_0), where Φl=π(2kl)−1/2Γ(12+ν)Γ(12−ν)\Phi_l=\sqrt{\pi}(2k_l)^{-1/2}\Gamma(\frac{1}{2}+\nu)\Gamma(\frac{1}{2}-\nu) is the integral ∫0∞dt t−1/2e−kltKν(klt)\int_0^\infty dt\,t^{-1/2}e^{-k_lt}K_{\nu}(k_lt) of the cut line over its mixing vertex.

With s=sin⁡πνs=\sin\pi\nu, the cutting rule then gives [A]

A3=−2ψ3+2π(1−s)[∑lΦlψ2[l]+2sπ∑lΦmΦnψ1[l]−2π2(1−s)(1+2s)Φ1Φ2Φ3ψ0].(5)A_3=-2\psi_3+\frac{2}{\pi}(1-s)\left[\sum_l\Phi_l\psi_2[l]+\frac{2s}{\pi}\sum_l\Phi_m\Phi_n\psi_1[l]-\frac{2}{\pi^2}(1-s)(1+2s)\Phi_1\Phi_2\Phi_3\psi_0\right]. \tag*{(5)}

With two time orderings per uncut line, tl<t0t_l<t_0 or tl>t0t_l>t_0, ψ3\psi_3, ψ2[l]\psi_2[l], ψ1[l]\psi_1[l] and ψ0\psi_0 give 15 integrals that we compute, while the others follow by permuting the legs.

**Two routes to closed forms—**To decouple the two times in KνIνK_{\nu}I_{\nu}, we replace every Bessel function by its Poisson integral [47],

Iν(z)=(z/2)νπΓ(ν+12)∫−11dx (1−x2)ν−12ezx,I_{\nu}(z)=\frac{(z/2)^{\nu}}{\sqrt{\pi}\Gamma(\nu+\frac{1}{2})}\int_{-1}^{1}dx\,(1-x^2)^{\nu-\frac{1}{2}}e^{zx},
Kν(z)=π(z/2)νΓ(ν+12)∫1∞dλ (λ2−1)ν−12e−zλ,(6)K_{\nu}(z)=\frac{\sqrt{\pi}(z/2)^{\nu}}{\Gamma(\nu+\frac{1}{2})}\int_{1}^{\infty}d\lambda\,(\lambda^2-1)^{\nu-\frac{1}{2}}e^{-z\lambda}, \tag*{(6)}

so that in a fixed time ordering, every exponent is linear in the times. After the rescaling tl=t0wlt_l=t_0w_l, the integral over t0t_0 is elementary and leaves a power of E=∑lklξlE=\sum_l k_l\xi_l, with ξl\xi_l linear in each of the auxiliary variables wlw_l, χl\chi_l and λl\lambda_l. (For related Euler-type representations, see [48, 49].) To integrate over the auxiliary variables, we separate the terms of EE with the MB representation of a power of a sum [50]. The auxiliary integrals then factorize into Beta functions, and every ordering becomes a multifold MB integral over a ratio of Gamma functions and powers of the momentum ratios.

The mass enters only through ν\nu in the Gamma functions. With ν=12−δ\nu=\frac{1}{2}-\delta near the conformally coupled mass, m2=2m^2=2, and ν=32−δ\nu=\frac{3}{2}-\delta in the massless limit, m2=3δ−δ2m^2=3\delta-\delta^2, δ\delta enters linearly, like the regulator of dimensional regularization. However, as δ→0\delta\to0, poles of the Gamma functions on opposite sides of a contour merge and pinch it, and the integrals must first be resolved [51, 52, 53, 54]. To evaluate them, we use our package MBAnayltica [55], which combines the integration method of angular phase-space integrals [56, 57] with geometric methods for multifold MB integrals [58, 59, 60, 61] originally developed for Feynman integrals [52, 62]. It resolves the δ\delta-singularities of all 15 integrals, rewrites them as Euler-type integrals with linear denominators, and integrates their coefficients in δ\delta into MPLs [38, 63] of the momentum ratios.

A second route starts from a conformally coupled σ\sigma, ν=12\nu=\frac{1}{2}, whose kernel is elementary,

K12(kt>)I12(kt<)=e−k∣t−t′∣−e−k(t+t′)2ktt′,(7)K_{\frac{1}{2}}(kt_>)I_{\frac{1}{2}}(kt_<)=\frac{e^{-k\lvert t-t'\rvert}-e^{-k(t+t')}}{2k\sqrt{tt'}}, \tag*{(7)}

a flat-space propagator [64] rescaled by 1/tt′1/\sqrt{tt'}. Once we write the inverse powers of the times as integrals over shifts ω\omega of the energies, t−n−1=∫0∞dω ωne−ωt/n!t^{-n-1}=\int_0^\infty d\omega\,\omega^n e^{-\omega t}/n!, the time integrals of every ordering are those of a flat-space tree, and every ψn\psi_n becomes an integral, over the shifts, of the flat-space wavefunction of the tree [64], a rational function with denominators linear in the energies. Away from ν=12\nu= \frac{1}{2}, each order in δ\delta adds a mass insertion, a quadratic vertex proportional to 2−m22-m^2 with measure dt/t2dt/t^2, on an uncut line, or a factor log⁡(λ2−1)\log(\lambda^2-1) under the Poisson integral of a cut line [A], which preserves this structure. We evaluate the resulting Euler-type integrals as in the MB route, and the same steps apply in the massless limit, where the kernel is also elementary.

The two routes agree where they overlap and together give all coefficients below. We also checked every ψn\psi_n by direct numerical integration.

**Near the conformally coupled mass—**With ν=12−δ\nu=\frac{1}{2}-\delta, the triple exchange is regular at δ=0\delta=0, and we obtain the Taylor coefficients A3(0)A_3^{(0)}, A3(1)A_3^{(1)} and A3(2)A_3^{(2)} of A3=∑n≥0A3(n)δnA_3=\sum_{n\geq0}A_3^{(n)}\delta^n for general kinematics as combinations of MPLs with rational coefficients [65]. Although A3A_3 is regular, the factors in Eq. (5) are not. Indeed, the factor 1−s1-s vanishes at δ=0\delta=0, while Φl\Phi_l and ψ0\psi_0 diverge. Expanding every factor in δ\delta first, we find

A3(0)=−π48k1k2k3+∑lπ22kmknI1[l]−2ψ3(0),(8)A_3^{(0)}=-\frac{\pi^4}{8k_1k_2k_3}+\sum_l\frac{\pi^2}{2k_mk_n}I_1[l]-2\psi_3^{(0)}, \tag*{(8)}

where the three terms come from ψ0\psi_0, ψ1[l]\psi_1[l] and ψ3\psi_3, respectively. The function I1[l]=(2/π)kmkn ψ1(0)[l]I_1[l]=(2/\pi)\sqrt{k_mk_n}\,\psi_1^{(0)}[l] has transcendental weight two [see Eq. (B1)], and ψ3(0)\psi_3^{(0)}, a combination of 91 MPLs, has weight four.

More generally, the coefficients of k1k2k3A3k_1k_2k_3A_3 at δ0\delta^0, δ1\delta^1 and δ2\delta^2 are pure functions [66], with rational coefficients and uniform transcendental weight four, five and six. Their symbols [67], which keep track of the arguments, or letters, of the iterated integrals and their ordering, use the eight-letter alphabet {2,ki,kT,kT−2ki}\{2,k_i,k_T,k_T-2k_i\} of conformally coupled fields [68, 69], with kT=k1+k2+k3k_T=k_1+k_2+k_3. Hence, a coefficient can be singular only where a letter vanishes, and each such point has a physical meaning. The point ki=0k_i=0 is the squeezed limit, and kT=0k_T=0, reached only in the unphysical region, is dominated by early times and probes the flat-space scattering amplitude [64]. The points ki=km+knk_i=k_m+k_n are the folded configurations, where a particle could decay into two collinear ones, which requires a particle in the initial state, absent in the Bunch–Davies vacuum [18]. Accordingly, the letters kT−2kik_T-2k_i never appear as leftmost entries of the symbol, which encode the branch points, and the functions have branch points only at ki=0k_i=0 and kT=0k_T=0.

This pattern has a simple explanation. Each massive line is δ\delta times a twisted dlog⁡d\log integral [70] [see Eq. (B2)], and every vertex has the measure dt/tdt/t. This predicts weight n+4n+4 at order δn\delta^n, and V+nV+n for a tree with VV such vertices, as anticipated, for the single exchange, V=2V=2, in [70]. The argument is heuristic, but we have checked it [B], also for two more topologies.

**Light fields—**In the massless limit, with ν=32−δ\nu=\frac{3}{2}-\delta, A3A_3 instead has a pole of fourth order at δ=0\delta=0, and we compute all coefficients up to the finite part. The pole arises because a mode of a light field decays only as e−δNe^{-\delta N} during the NN e-folds it spends outside the Hubble radius 1/H1/H [3, 17], so that each of the four time integrals gives a factor 1/δ1/\delta. Solving the equations of motion in this limit [C], we obtain the leading term, the local shape [71]

SΠ=ρ3μp34e3δ4,S_{\Pi}=\frac{\rho^3\mu p_3}{4e_3\delta^4},

with p3=∑iki3p_3=\sum_i k_i^3 and e3=k1k2k3e_3=k_1k_2k_3. With the elementary symmetric polynomials e1=kTe_1=k_T, e2=k1k2+k2k3+k3k1e_2=k_1k_2+k_2k_3+k_3k_1 and e3e_3, we write the full expansion for light fields as

−3e32A3=Tδ∑n=04qnδn−4+O(δ),(9)-3e_3^2A_3=\mathcal{T}_{\delta}\sum_{n=0}^{4}q_n\delta^{n-4}+O(\delta), \tag*{(9)}
Tδ=∑jkj3∏i≠j(2kikT)δ,\mathcal{T}_{\delta}=\sum_j k_j^3\prod_{i\ne j}\left(\frac{2k_i}{k_T}\right)^\delta,

where q0=1q_0=1 and q1=(−p3+2e1e2−8e3)/p3q_1=(-p_3+2e_1e_2-8e_3)/p_3 are rational [C], and q2q_2, q3q_3 and q4q_4 are MPLs. In the quasi-local shape [3, 17] Tδ\mathcal{T}_{\delta}, each term kj3k_j^3 comes from the power spectra of the other two legs, proportional to (kikl)−3(k_ik_l)^{-3}, and carries one factor (2ki/kT)δ(2k_i/k_T)^\delta for each of them, the decay of σ\sigma between the horizon crossing of that mode and the cubic vertex. This factor reproduces the exact rescaling in the squeezed limit [3, 4], SΠ∝(k3/k1)−1+δS_{\Pi}\propto(k_3/k_1)^{-1+\delta}, and resums the leading logarithms, so that q2q_2, q3q_3 and q4q_4 remain finite as k3→0k_3\to0. In the equilateral configuration, A3(1,1,1;δ)=∑n=−40cnδn+O(δ)A_3(1,1,1;\delta)=\sum_{n=-4}^{0}c_n\delta^n+O(\delta), with

c−4=−1,c−3=−73+2L,L=log⁡32,c_{-4}=-1,\qquad c_{-3}=-\frac{7}{3}+2L,\qquad L=\log\frac{3}{2},
c−2=−73+6L−2L2+π24−Li⁡213.(10)c_{-2}=-\frac{7}{3}+6L-2L^2+\frac{\pi^2}{4}-\operatorname{Li}_2\frac{1}{3}. \tag*{(10)}

We obtained c−4c_{-4} through c−1c_{-1} (see [C]) by direct integration, and c0c_0 [see [65]] using the PSLQ algorithm [72].

The pole is regulated by the duration of inflation [3, 73]. If inflation ends NN e-folds after the horizon crossing of the largest momentum, which we set to one, each of the four time integrals gives (1−e−δN)/δ(1-e^{-\delta N})/\delta instead of 1/δ1/\delta. Matching the super-horizon evolution to the propagators at horizon crossing [C], we find, for δ→0\delta\to0 at fixed δN\delta N,

A3(N)A3(∞)=(1−e−δN)4[1+(4ΔN−Λ)δeδN−1+O(δ2)],(11)\frac{A_3(N)}{A_3(\infty)}=(1-e^{-\delta N})^4\left[1+\frac{(4\Delta N-\Lambda)\delta}{e^{\delta N}-1}+O(\delta^2)\right], \tag*{(11)}

with elementary functions ΔN\Delta N and Λ\Lambda of the momenta, ΔN=0.86\Delta N=0.86 and Λ=0\Lambda=0 in the equilateral configuration. A massless field gives a bispectrum that grows as N4N^4 [3], and the massless and late-time limits do not commute.

**The non-Gaussian amplitude—**For observations, we use the shape function of ζ\zeta [43], Sζ≡(k1k2k3)2Bζ/(2πΔζ)4=−SΠ/(2πΔζ)S_\zeta\equiv(k_1k_2k_3)^2B_\zeta/(2\pi\Delta_\zeta)^4=-S_{\Pi}/(2\pi\Delta_\zeta), with BζB_\zeta defined like BπB_\pi and Δζ2\Delta_\zeta^2 the dimensionless power spectrum. In the equilateral configuration, Sζ=910fNLS_\zeta=\frac{9}{10}f_{\mathrm{NL}}, so that, restoring HH,

fNL(m)=56gA3(1,1,1;δ),g=ρ3μ2πΔζH4,(12)f_{\mathrm{NL}}(m)=\frac{5}{6}gA_3(1,1,1;\delta),\qquad g=\frac{\rho^3\mu}{2\pi\Delta_\zeta H^4}, \tag*{(12)}

with δ=32−(94−m2H2)1/2\delta= \frac{3}{2} - \left(\frac{9}{4} - \frac{m^2}{H^2}\right)^{1/2}. For m≪Hm \ll H, fNL≃−56g(3H2m2)4f_{\mathrm{NL}} \simeq-\frac{5}{6}g\left(\frac{3H^2}{m^2}\right)^4, and at fixed gg the signal grows as m−8m^{-8}, from fNL/g=−48f_{\mathrm{NL}}/g=-48 at m=1.02Hm=1.02H to −1.4⋅105-1.4\cdot10^5 at 0.38H0.38H. Equation (12) applies at weak mixing, ρ≪δH\rho\ll\delta H [3], and at late times. We test Eqs. (2) and (12) against CosmoFlow [25, 41], which computes the correlators of the theory of Eq. (1) numerically, without a mass expansion. At m=0.38Hm=0.38H, for instance, our expansion for light fields agrees with CosmoFlow to 3⋅10−63\cdot10^{-6} in all tested triangles. Similarly, at a finite duration, Eq. (11) agrees with it within 2.1%2.1\% at m=0.27Hm=0.27H, 0.38H0.38H and 0.54H0.54H for 10≤N≤6010\leq N\leq60. For N=60N=60, the amplitude at m=0.38Hm=0.38H is 82%82\% of its late-time value.

*One function of the mass—*At fixed positive momenta, A3A_3 is an even meromorphic function of ν\nu, whose possible poles, as we show in [D], lie only at ν=±(32+J)\nu=\pm\left(\frac{3}{2}+J\right), J=0,1,…J=0,1,\ldots, that is, at m2/H2=2−(J+1)(J+2)=0,−4,−10,…m^2/H^2=2-(J+1)(J+2)=0,-4,-10,\ldots (Figure 3). In particular, A3A_3 has no singularity at positive m2m^2. The poles closest to the conformally coupled mass are the massless ones, ν=±32\nu=\pm\frac{3}{2}, with the Laurent coefficients c−4L,…,c−1Lc^L_{-4},\ldots,c^L_{-1} of Eq. (9) for the same triangle (the cnc_n of Eq. (10) in the equilateral configuration). Subtracting them leaves a function that is regular up to the next pole, at m2=−4H2m^2=-4H^2, so that the series in

Figure 3. Analytic structure of A3A_3 in the complex plane of m2/H2m^2/H^2. Crosses mark possible poles, of order up to four (bold) or simple (thin). Dashed circles show the convergence of the series in Eq. (13) as further poles are subtracted. The lower strip magnifies the box, with the threshold m=32Hm=\frac{3}{2}H (orange) and the zero of the equilateral fNLf_{\mathrm{NL}} (open circle).

A3=∑j=04c−jL[(32−ν)−j+(32+ν)−j]+∑n≥0βnun,(13)A_3=\sum_{j=0}^{4}c^L_{-j}\left[\left(\frac{3}{2}-\nu\right)^{-j}+\left(\frac{3}{2}+\nu\right)^{-j}\right]+\sum_{n\geq0}\beta_n u^n, \tag*{(13)}

with u=ν2−14=2−m2/H2u=\nu^2-\frac{1}{4}=2-m^2/H^2, converges for ∣u∣<6|u|<6, which includes all light masses. With β0\beta_0, β1\beta_1 and β2\beta_2 from our closed forms near the conformally coupled mass, Eq. (13) gives fNL(m)f_{\mathrm{NL}}(m) from the massless limit to 32H\frac{3}{2}H, within 4⋅10−44\cdot10^{-4} of CosmoFlow at the sampled masses. Once we also subtract the poles at ν=52\nu=\frac{5}{2} and 72\frac{7}{2} and sum numerical coefficients of the expansion in δ=12−ν\delta=\frac{1}{2}-\nu through δ13\delta^{13} with Padé approximants, we continue it to heavy fields, in agreement with CosmoFlow to 1.1⋅10−61.1\cdot10^{-6} and with the Laplace-space solution [35] up to 3H3H, away from m=2.409Hm=2.409H, where fNLf_{\mathrm{NL}} changes sign [D]. Furthermore, expanded numerically around both limits, the series of [36], exact in the mass, reproduce our coefficients.

*Templates and constraints—*To calibrate the templates, we first take the squeezed limit of Eq. (9),

m/Hm/HlocalQSFequil.orth.fNLf_{\mathrm{NL}}κ\kappa
0.380.9950.9810.35−0.57-0.57−1.4±6.7-1.4\pm6.71.82
0.540.9840.9820.40−0.60-0.60−1.7±8.1-1.7\pm8.11.67
0.750.9420.9850.49−0.66-0.66−2.0±10.8-2.0\pm10.81.46
1.020.8150.9920.64−0.71-0.71−0.9±16.2-0.9\pm16.21.24

Table 1. Cosines rXY=FXY/FXXFYYr_{XY}=F_{XY}/\sqrt{F_{XX}F_{YY}} of the analytic shape (E) with the templates, for the Planck 2018 temperature and polarization Fisher matrix FF of CMB-BEST [74], and Planck constraints on fNL(m)f_{\mathrm{NL}}(m) for E (68% CL). The QSF template (Q) matches E up to the rescaling κ=FQE/FQQ\kappa=F_{QE}/F_{QQ}. The masses are those of δ=32−ν=0.05\delta=\frac{3}{2}-\nu=0.05, 0.1, 0.2 and 0.4.

in which the qnq_n reduce to zeta values, and qsq(δ)=∑nqn(1,1,0)δnq_{\mathrm{sq}}(\delta)=\sum_n q_n(1,1,0)\delta^n is

qsq=1+δ+(1−π212)δ2+(7−π212−72ζ(3))δ3+(27−3π24−52ζ(3)−π46)δ4+O(δ5),(14)q_{\mathrm{sq}}=1+\delta+\left(1-\frac{\pi^2}{12}\right)\delta^2+\left(7-\frac{\pi^2}{12}-\frac{7}{2}\zeta(3)\right)\delta^3 +\left(27-\frac{3\pi^2}{4}-\frac{5}{2}\zeta(3)-\frac{\pi^4}{6}\right)\delta^4+O(\delta^5), \tag*{(14)}

with coefficients that we identified using PSLQ. They agree with the weak-mixing limit of the squeezed coefficient of [32], which we continued analytically to light fields. The QSF template [3] used in the Planck analyses [8, 9] and in forecasts for galaxy surveys [75] has the same scaling in the squeezed limit, (k3/k1)−1+δ(k_3/k_1)^{-1+\delta}, but, at equal equilateral amplitude, a squeezed bispectrum lower by factors of 2.35 for m→0m\to0 and 1.57 at 1.02H1.02H (Figure 4). As a result, an estimator built on this template returns κfNL\kappa f_{\mathrm{NL}} when the data contain our analytic shape with amplitude fNLf_{\mathrm{NL}}, with κ=1.82\kappa=1.82 at m=0.38Hm=0.38H and 1.24 at 1.02H1.02H (Table I). An uncalibrated bound on the coupling gg is therefore too weak by the same factor. With cosines above 0.98, the calibrated template has a Fisher error at most 2% larger than that of the analytic shape.

Figure 4. Shape functions along isosceles triangles, k1=k2k_1=k_2. In the squeezed limit, the QSF template falls below our closed form for light fields, which agrees with CosmoFlow.

Truncating the sum in Eq. (9) after q1q_1 and normalizing to ST(k,k,k)=910fNLS_{\mathrm{T}}(k,k,k)=\frac{9}{10}f_{\mathrm{NL}}, we obtain a new template,

ST=3fNL10(3/2)2δ1+73δTδk1k2k3(1+q1δ),(15)S_{\mathrm{T}}=\frac{3f_{\mathrm{NL}}}{10}\frac{(3/2)^{2\delta}}{1+\frac{7}{3}\delta}\frac{\mathcal{T}_{\delta}}{k_1k_2k_3}(1+q_1\delta), \tag*{(15)}

an elementary function with two parameters, the amplitude fNLf_{\mathrm{NL}} and, through δ\delta, the mass. Its momentum dependence, chiefly the slope in the squeezed limit, measures the mass, and its amplitude then gives the coupling through Eq. (12). Fitted to the analytic shape at the masses of Table I, it recovers the amplitude within 5% and the mass within 0.002H0.002H. In contrast, the QSF template, fitted in the same way, prefers lighter fields, with m→0m\to0 for 0.38H0.38H and 0.54H0.54H, 0.52H0.52H for 0.75H0.75H and 0.89H0.89H for $1.02H, a bias larger than the statistical error once the signal-to-noise ratio exceeds 6 to 9.

Using the CMB-BEST pipeline [74], we reproduce the Planck template constraints [76] and obtain those of the analytic shape (Table I). The corresponding 95% CL bound on ∣g∣|g| ranges from 1.0⋅10−41.0\cdot10^{-4} at m=0.38Hm=0.38H to 0.7 at 1.02H1.02H. In a simplified Fisher analysis, the Simons Observatory [15] and a CMB-S4-like benchmark [77], combined with Planck, would reduce the error on fNLf_{\mathrm{NL}} by factors of 1.6 and 2.1. In galaxy surveys, a light field also generates a scale-dependent correction to the linear galaxy bias [75, 78, 79], Δb(k)∝fNLsqkδ−2/T(k)\Delta b(k)\propto f_{\mathrm{NL}}^{\mathrm{sq}}k^{\delta-2}/T(k), with T(k)T(k) the matter transfer function. Its slope measures the mass, and, in the normalization of local non-Gaussianity [71], Eq. (14) gives its amplitude, fNLsq(m)=−56g qsq(δ)/δ4f_{\mathrm{NL}}^{\mathrm{sq}}(m)=-\frac{5}{6}g\,q_{\mathrm{sq}}(\delta)/\delta^4, the calibrated theoretical input for SPHEREx [16] and other surveys [80].

*Mixing and self-coupling—*The bispectrum constrains only one combination of the mixing and the self-coupling. In terms of ρˉ=ρ/(δH)\bar{\rho} = \rho/(\delta H) and μˉ=μ/(2πδH)\bar{\mu} = \mu/(2\pi\delta H), both small at weak coupling [C], the coupling is g=ρˉ3μˉ4δ/Δζg = \bar{\rho}^{3}\bar{\mu}^{4}\delta/\Delta_{\zeta}, and the Planck bounds on gg give ρˉ3μˉ<7.5⋅10−4\bar{\rho}^{3}\bar{\mu} < 7.5 \cdot10^{-4} at m=0.38Hm = 0.38H and 1.2⋅10−31.2 \cdot10^{-3} at 1.02H1.02H. These bounds therefore correspond to ρˉ=μˉ≈0.17\bar{\rho} = \bar{\mu} \approx0.17 at m=0.38Hm = 0.38H and 0.190.19 at 1.02H1.02H, inside the weakly coupled regime. At weak mixing, the bispectrum contains little information about the self-coupling, since the bound on μˉ\bar{\mu} grows as 1/ρˉ31/\bar{\rho}^{3}.

A second observable bounds a different combination. It is the trispectrum from the exchange of σ\sigma between two cubic vertices [5, 81], which depends on ρ4μ2\rho^{4}\mu^{2}. In its collapsed limit, two pairs of short modes share one long mode of σ\sigma. In the squeezed bispectrum, the same long mode appears only through its correlation with the inflaton, of magnitude ϵ=πρ/[2Hsin⁡(π4+πν2)]\epsilon= \pi\rho/[2H\sin(\frac{\pi}{4} + \frac{\pi\nu}{2})] [5] [C]. With Eq. (14), the leading collapsed amplitude, in the normalization of [39], is therefore

τNLlight=(65fNL)2ϵ2=[ρˉ2μˉqsq(δ)Δζsin⁡(πδ/2)πδ/2]2.(16)\tau_{\mathrm{NL}}^{\mathrm{light}} = \frac{\left(\frac{6}{5} f_{\mathrm{NL}}\right)^2}{\epsilon^2} = \left[\frac{\bar{\rho}^{2}\bar{\mu} q_{\mathrm{sq}}(\delta)}{\Delta_{\zeta}} \frac{\sin(\pi\delta/2)}{\pi\delta/2}\right]^2 . \tag*{(16)}

For light fields, where ϵ≃ρˉ\epsilon\simeq\bar{\rho}, the trispectrum is enhanced by 1/ρˉ21/\bar{\rho}^{2} relative to the squared bispectrum, and it bounds ρˉ2μˉ\bar{\rho}^{2}\bar{\mu}.

The Planck analysis of the collapsed trispectrum [39] gives τNLlight<2360\tau_{\mathrm{NL}}^{\mathrm{light}} < 2360 for a massless field (95% CL), and its result at m=1.12Hm = 1.12H corresponds to 1.6⋅1051.6 \cdot10^{5}. Interpolating in the mass [C], we estimate ρˉ2μˉ<2.7⋅10−3\bar{\rho}^{2}\bar{\mu} < 2.7 \cdot10^{-3} at m=0.38Hm = 0.38H and 9.0⋅10−39.0 \cdot10^{-3} at 1.02H1.02H. Figure 2 shows both bounds as upper limits on μˉ\bar{\mu} at a given ρˉ\bar{\rho}. They fall as 1/ρˉ31/\bar{\rho}^{3} and 1/ρˉ21/\bar{\rho}^{2}, and cross at ρˉ≈0.28\bar{\rho} \approx0.28 for m=0.38Hm = 0.38H and at 0.140.14 for 1.02H1.02H, where the bound is already μˉ≈0.5\bar{\mu} \approx0.5. At weaker mixing, we note that the trispectrum gives the stronger bound.

For instance, at ρˉ=0.1\bar{\rho} = 0.1 and m=0.38Hm = 0.38H, it lowers the bound from μˉ<0.75\bar{\mu} < 0.75 to 0.270.27. If both signals were detected, the ratio of the trispectrum to the squared bispectrum, 1/ϵ21/\epsilon^{2}, would therefore measure the mixing alone, and the bispectrum would then give the self-coupling. In galaxy surveys, the same enhancement also appears as a stochastic bias [82].

**Conclusions—**With methods from Feynman integrals, we have obtained, for the first time, the QSF triple exchange of light fields at weak mixing in closed form in the momenta, as one meromorphic function of the mass that agrees with the series exact in the mass [36]. The exact shape calibrates the QSF template of the Planck analyses and yields a simple template for the coupling and the mass. With the collapsed trispectrum, it can separate the mixing from the self-coupling. Our methods extend to the double exchange, to complete trispectra and to higher orders in the mass expansion.

*Note added—*During the final preparation of this Letter, the triple exchange was expressed as a convergent double series at any mass and mixing strength [36]. We checked that our results numerically agree with theirs.

*Acknowledgments—*The work of TA is supported by the Deutsche Forschungsgemeinschaft (DFG) through Research Unit FOR2926, Next Generation Perturbative QCD for Hadron Structure: Preparing for the Electron-Ion Collider, project number 409656131. The work of SMH is supported by the Academic Research Sabbatical Program (ARSP) of the University of Regensburg. SB gratefully acknowledges the financial support of the Swiss National Science Foundation Project No. P500-2_235325 and of the US Department of Energy under contract DE-AC02-76SF00515.

† [email protected]

‡ [email protected]

§ [email protected]

References

End Matter

Cutting rule and computational details

Here we derive the integrals of Eqs. (3) and (4) and the coefficients of Eq. (5). In the Euclidean time integrals, a vertex vv carries tvrv−1e−xvtvt_v^{r_v-1}e^{-x_vt_v}, with xvx_v the energy of its external legs. The power of tvt_v collects the measure a4=t−4a^4=t^{-4}, a factor tv3/2t_v^{3/2} per line end and, at a mixing vertex, tv2t_v^2 from the π˙c\dot{\pi}_c leg, which also supplies e−kltve^{-k_lt_v}. Hence r0=32r_0=\frac{3}{2} and x0=0x_0=0 at the cubic vertex, and rl=12r_l=\frac{1}{2} and xl=klx_l=k_l at the mixing vertices, which gives Eqs. (3) and (4). The coefficients of Eq. (5) come from [28], where the correlator is a sum over the subsets CC of cut lines,

∑C(−1)3−nC2πnCRe⁡∏BψBF,(17)\sum_C(-1)^{3-n_C}\frac{2}{\pi^{n_C}}\operatorname{Re}\prod_B\psi_B^F , \tag*{(17)}

with nCn_C of the three lines cut and ψBF\psi_B^F the integral of each connected piece BB left by the cuts, evaluated on the uncut lines with the Feynman kernel

GνF=Kν(kt>)[Iν−iπeiπνKν](kt<).(18)G_\nu^F=K_\nu(kt_>)\left[I_\nu-\frac{i}{\pi}e^{i\pi\nu}K_\nu\right](kt_<). \tag*{(18)}

Its KνKνK_\nu K_\nu part factorizes as on a cut line, with the mixing-vertex integral Φl\Phi_l. Collecting the terms with nn time-ordered lines, we obtain their weight

∑j=03−n(3−nj)2(−1)3−jπ3−nRe⁡[(s−icos⁡πν)3−n−j],(19)\sum_{j=0}^{3-n}\binom{3-n}{j}\frac{2(-1)^{3-j}}{\pi^{3-n}}\operatorname{Re}\left[(s-i\cos\pi\nu)^{3-n-j}\right], \tag*{(19)}

which evaluates to the four coefficients of Eq. (5).

We compute all ψn\psi_n through the finite part of A3A_3 in the massless limit, from δ−4\delta^{-4} to δ0\delta^0, and through δ2\delta^2 near the conformally coupled mass. In the direct integration, an uncut line becomes a chain of conformal kernels, Eq. (7), joined by mass insertions, while a cut line is expanded in δ\delta under its Poisson integral, Eq. (6), which gives powers of log⁡(λ2−1)\log(\lambda^2-1). We checked that the assembled A3A_3 is symmetric under permutations of the legs to between 37 and 39 digits. We identified the rational coefficient q1q_1 by an ansatz fitted at 28 triangles, which then holds to 10−1510^{-15} at all 117 triangles we evaluated.

Structure near the conformally coupled mass

The simplest example of our closed forms is the weight-two function I1[l]I_1[l] of Eq. (8). In units k1=1k_1=1, it reads

klI1[l]=π212+12G(0,1;1−2kl)−12G(0,1;1−kT)−12G(11−kT,11−2kl;1)+14log⁡2kT.(20)k_l I_1[l]=\frac{\pi^2}{12}+\frac{1}{2}G(0,1;1-2k_l)-\frac{1}{2}G(0,1;1-k_T)-\frac{1}{2}G\left(\frac{1}{1-k_T},\frac{1}{1-2k_l};1\right)+\frac{1}{4}\log^2 k_T . \tag*{(20)}

The weights we find in the main text follow from the structure of the integrals. Each Bessel function of a massive line is δ\delta times a twisted dlog⁡d\log integral [70],

K12−δ(z)=−δπ 2δ−12Γ(1−δ)z−12−δ∫1∞(λ2−1)−δdlog⁡(λ2−1)e−zλ,(21)K_{\frac{1}{2}-\delta}(z)=-\delta\frac{\sqrt{\pi}\,2^{\delta-\frac{1}{2}}}{\Gamma(1-\delta)}z^{-\frac{1}{2}-\delta}\int_1^\infty(\lambda^2-1)^{-\delta}d\log(\lambda^2-1)e^{-z\lambda}, \tag*{(21)}

continued from δ<0\delta<0 (1l1−δ\frac{1}{l_1}-\delta likewise, with λ∈[−1,1]\lambda\in[-1,1]), and the vertices of Eq. (1) have rv=12nvr_v=\frac{1}{2}n_v, with nvn_v Bessel functions at the vertex. Every vertex measure is then a twisted dt/tdt/t, and the time integrals give the flat-space wavefunction [64] at shifted energies, a dlog⁡d\log form with constant coefficients. The coefficient of δn\delta^n is therefore pure of weight V+n=n+4V+n=n+4, with leftmost symbol entries at the physical branch points [18] and letters fixed by the δ\delta-independent singularities of the integrand. However, the argument is heuristic, since it assumes that the nested orderings, the boundary terms of the cut lines and the normalizations of Eq. (5) do not spoil this form. We therefore tested it with three checks. First, the functions, not only their symbols, are pure at δ0\delta^0, δ1\delta^1 and δ2\delta^2 (to 30 digits at three points). Second, every ψn\psi_n in Eq. (5) is pure on its own. Third, the single exchange, through fourth order, and the three-site chain with two massive lines, at its first two orders, have the predicted weights, n+2n+2 and n+3n+3. The argument does not apply to light fields, whose twist (λ2−1)1−δ(\lambda^2-1)^{1-\delta} is not a dlog⁡d\log, and indeed the coefficients in Eq. (14) do not have uniform weight.

Light fields and the regime of validity

The fourth-order pole has a simple explanation. With H=1H=1 and NN the number of ee-folds since horizon crossing, Eq. (1) gives, on super-horizon scales,

dπcdN=−ρσ,dσdN=−m23σ−μσ2,(22)\frac{d\pi_c}{dN}=-\rho\sigma,\qquad\frac{d\sigma}{dN}=-\frac{m^2}{3}\sigma-\mu\sigma^2, \tag*{(22)}

where we neglect d2σ/dN2d^2\sigma/dN^2 and use m23=δ+O(δ2)\frac{m^2}{3}=\delta+O(\delta^2). A light field decays only as σ1=σ∗e−δN\sigma_1=\sigma_*e^{-\delta N}, with σ∗\sigma_* its value at horizon crossing, and is converted into the curvature perturbation for 1/δ1/\delta ee-folds. It sources πc\pi_c at linear order and, through its self-interaction, at second order,

πc⊃−ρσ∗δ+ρμσ∗22δ2.(23)\pi_c\supset-\frac{\rho\sigma_*}{\delta}+\frac{\rho\mu\sigma_*^2}{2\delta^2}. \tag*{(23)}

With ⟨σ∗σ∗⟩′=1/(2k3)\langle\sigma_*\sigma_*\rangle'=1/(2k^3), where the prime denotes that (2π)3δ(3)(∑iki)(2\pi)^3\delta^{(3)}(\sum_i k_i) is dropped, Wick’s theorem gives the local shape

Sπ=ρ3μp34ϵ3δ4,(24)S_\pi=\frac{\rho^3\mu p_3}{4\epsilon^3\delta^4}, \tag*{(24)}

and hence the constant −3δ4-\frac{3}{\delta^4} of Eq. (2). In the equilateral configuration, with L=log⁡23L=\log\frac{2}{3} and ℓ=log⁡3\ell=\log3,

c−1=−313+6L−6L2+43L3+π24+π22L−(2L2+π22)ℓ−(3+L)Li⁡213+4Li⁡323+ζ(3)24.(25)c_{-1}=-\frac{31}{3}+6L-6L^2+\frac{4}{3}L^3+\frac{\pi^2}{4}+\frac{\pi^2}{2}L-(2L^2+\frac{\pi^2}{2})\ell-(3+L)\operatorname{Li}_2\frac{1}{3}+4\operatorname{Li}_3\frac{2}{3}+\frac{\zeta(3)}{24}. \tag*{(25)}

The same picture sets the regime of validity. The late-time results assume that inflation lasts much longer than the 1/δ1/\delta ee-folds over which σ\sigma decays. If it ends NN ee-folds after the horizon crossing of the largest momentum, set to one, we match the super-horizon evolution of the legs, with the conversion of σ\sigma into πc\pi_c accumulated until the end of inflation, to the massless integral over the cubic vertex at horizon crossing. The matching time cancels, which gives Eq. (11) with Λ=∑jkj3log⁡(kikl)/p3\Lambda=\sum_j k_j^3\log(k_i k_l)/p_3, with {i,l}\{i,l\} the legs other than jj, and ΔN=56−γE−12log⁡(2e1)+(4e13−9e1e2)/(6p3)\Delta N=\frac{5}{6}-\gamma_E-\frac{1}{2}\log(2e_1)+(4e_1^3-9e_1e_2)/(6p_3). Moreover, the matched amplitude reproduces the first two orders of Eq. (9) for δN≫1\delta N\gg1, including q1q_1, and gives −3ϵ3A3=p3[N4+(4ΔN−Λ)N3+O(N2)]-3\epsilon_3A_3=p_3[N^4+(4\Delta N-\Lambda)N^3+O(N^2)] for a massless field. In the squeezed limit, where Eq. (11) is not uniform, we solve Eq. (22) to the end of inflation, with a common shift of the horizon-crossing times fitted in the equilateral configuration, and reproduce CosmoFlow to 0.5% for N≥40N\ge40. For N=60N=60 and m≥0.4Hm\ge0.4H, a finite duration changes the shape by at most 1.5% for k3/k1≥10−3k_3/k_1\ge10^{-3}, but by 3.7% at m=0.30Hm=0.30H, where NN must be marginalized. Finally, the mixing is weak, and the cubic term μσ2\mu\sigma^2 is a correction to the linear one, δσ\delta\sigma, at the typical amplitude σ∼H/2π\sigma\sim H/2\pi, for

ρˉ=ρδH≪1,μˉ=μ2πδH≪1.(26)\bar{\rho}=\frac{\rho}{\delta H}\ll1,\qquad\bar{\mu}=\frac{\mu}{2\pi\delta H}\ll1. \tag*{(26)}

The linear term of Eq. (23) also gives the magnitude of the correlation of πc\pi_c with σ\sigma at long wavelengths, ϵ≃ρ/δ\epsilon\simeq\rho/\delta, and the late-time limit of the mixed propagator [27] gives ϵ=πρ/[2sin⁡(π4+πν2)]\epsilon=\pi\rho/[2\sin(\frac{\pi}{4}+\frac{\pi\nu}{2})] at any light mass. For the trispectrum, we identify the leading collapsed term with the template of [39] and neglect subleading terms. At ν=32\nu=\frac{3}{2} we use its 95% limit for the local shape, with which the template coincides. At ν=1,12\nu=1,\frac{1}{2} and 00, we turn its values into 95% limits with a Gaussian likelihood and a positive flat prior. We take the weaker pipeline and interpolate the logarithm of the limit in ν\nu with a cubic polynomial (a linear interpolation changes the bounds by less than 7%). Equation (14) is accurate to 1.3% at δ=0.4\delta=0.4.

Poles in the mass and the sign change

The poles of A3A_3 in the mass arise at late times. The bispectrum is one time integral over the mixed propagators of the three legs [27]. The Bunch–Davies initial condition of the propagators makes its early-time part holomorphic in ν\nu, while the late-time powers apa^p of the legs integrate to ap+1/(p+1)a^{p+1}/(p+1), with denominators that vanish only at ν=±(32+J)\nu=\pm(\frac{3}{2}+J), all at m2≤0m^2\le0. Since A3A_3 is even in ν\nu, it is a meromorphic function of ν\nu. At ν=32+J\nu=\frac{3}{2}+J with even JJ, the triple and double cuts contribute and the poles are of order at most four. For odd JJ, only the double cuts survive and the poles are at most simple, with

Res⁡ν=5/2A3=−π232∑lkl[3(km2+kn2)−kl2](kmkn)3.(27)\operatorname*{Res}_{\nu=5/2} A_3=-\frac{\pi^2}{32}\sum_l\frac{k_l\left[3(k_m^2+k_n^2)-k_l^2\right]}{(k_m k_n)^3}. \tag*{(27)}

For m≫Hm\gg H, we integrate out the field and find A3≃4H6/(m6kT3)A_3\simeq4H^6/(m^6k_T^3), in agreement with the equilateral limit of [36]. For positive gg, the amplitude is thus negative for light fields and positive for heavy ones, and it must vanish in between, at m=2.409Hm=2.409H in the equilateral configuration (Figure 3).

Origin of the template bias

In the squeezed limit,

Sζ(1,1,r)→910fNLCr1+δ,r=k3k1→0,(28)S_\zeta(1,1,r)\to\frac{9}{10}f_{\mathrm{NL}}Cr^{1+\delta},\qquad r=\frac{k_3}{k_1}\to0, \tag*{(28)}

where the slope measures the mass and CC fixes the ratio of the squeezed to equilateral signal. For m→0m\to0, the exact shape is local, with C=23C=\frac{2}{3}, whereas the QSF template [3],

S∝x1/2Nν(3x),x=k1k2k3kT3,(29)S\propto x^{1/2}N_\nu(3x),\qquad x=\frac{k_1k_2k_3}{k_T^3}, \tag*{(29)}

with NνN_\nu the Neumann function, has CQSF→0.284C_{\mathrm{QSF}}\to0.284. At m=1.02Hm=1.02H, C=0.705C=0.705 against $C_{\mathrm{QSF}}=0.449 ({xref}Fig. 4`). The template thus has the correct slope but too small a squeezed signal, the origin of both κ\kappa and the bias toward lighter fields. Consequently, in galaxy surveys, which measure fNLsq=23CfNLf_{\mathrm{NL}}^{\mathrm{sq}}=\frac{2}{3}C f_{\mathrm{NL}}, it overestimates the coupling by C/CQSF=1.6C/C_{\mathrm{QSF}}=1.6 to 2.2.

References

  1. [1]Y. Akrami et al. (Planck), Planck 2018 results. X. Constraints on inflation, Astron. Astrophys. 641, A10 (2020), arXiv:1807.06211 [astro-ph.CO].DOI
  2. [2]J. M. Maldacena, Non-Gaussian features of primordial fluctuations in single field inflationary models, JHEP 2003 (05), 013, arXiv:astro-ph/0210603.DOI
  3. [3]X. Chen and Y. Wang, Quasi-Single Field Inflation and Non-Gaussianities, JCAP 2010 (04), 027, arXiv:0911.3380 [hep-th].DOI
  4. [4]D. Baumann and D. Green, Signatures of Supersymmetry from the Early Universe, Phys. Rev. D 85, 103520 (2012), arXiv:1109.0292 [hep-th].DOI
  5. [5]V. Assassi, D. Baumann, and D. Green, On Soft Limits of Inflationary Correlation Functions, JCAP 2012 (11), 047, arXiv:1204.4207 [hep-th].DOI
  6. [6]T. Noumi, M. Yamaguchi, and D. Yokoyama, Effective field theory approach to quasi-single field inflation and effects of heavy fields, JHEP 2013 (06), 051, [7] N. Arkani-Hamed and J. Maldacena, Cosmological Collider Physics (2015), arXiv:1503.08043 [hep-th].
  7. [8]P. A. R. Ade et al. (Planck), Planck 2013 Results. XXIV. Constraints on primordial non-Gaussianity, Astron. Astrophys. 571, A24 (2014), arXiv:1303.5084 [astro-ph.CO].
  8. [9]W. Sohn, D.-G. Wang, J. R. Fergusson, and E. P. S. Shellard, Searching for cosmological collider in the Planck CMB data, JCAP 2024 (09), 016, arXiv:2404.07203 [astro-ph.CO].DOI
  9. [10]P. Suman, D.-G. Wang, W. Sohn, J. R. Fergusson, and E. P. S. Shellard, Searching for Cosmological Collider in the Planck CMB Data II: collider templates and Modal analysis (2025), arXiv:2512.22085 [astro-ph.CO].arxiv.org/abs/2512.22085
  10. [11]S. Kumar, Q. Lu, Z.-Z. Xianyu, and Y. Zhang, Scalars at the Cosmological Collider: Full Shapes of Tree Diagrams and Bispectrum Searches using Planck Data (2026), arXiv:2604.07434 [hep-ph].arxiv.org/abs/2604.07434
  11. [12]O. H. E. Philcox, Dissecting the Scalar Cosmological Collider with the Cosmic Microwave Background (2026), arXiv:2607.18369 [astro-ph.CO].arxiv.org/abs/2607.18369
  12. [13]G. Cabass, O. H. E. Philcox, M. M. Ivanov, K. Akitsu, S.-F. Chen, M. Simonović, and M. Zaldarriaga, BOSS constraints on massive particles during inflation: The cosmological collider in action, Phys. Rev. D 111, 063510 (2025), arXiv:2404.01894 [astro-ph.CO].arxiv.org/abs/2404.01894
  13. [14]D. Green, Y. Guo, J. Han, and B. Wallisch, Light fields during inflation from BOSS and future galaxy surveys, JCAP 2024 (05), 090, arXiv:2311.04482 [astro-ph.CO].
  14. [15]P. Ade et al. (Simons Observatory), The Simons Observatory: Science goals and forecasts, JCAP 2019 (02), 056, arXiv:1807.04475 [astro-ph.CO].
  15. [16]O. Doré et al. (SPHEREx), Cosmology with the SPHEREx All-Sky Spectral Survey (2014), arXiv:1412.4872 [astro-ph.CO].DOI
  16. [17]X. Chen and Y. Wang, Large non-Gaussianities with Intermediate Shapes from Quasi-Single Field Inflation, Phys. Rev. D 81, 063511 (2010), arXiv:0909.0496 [astro-ph.CO].DOI
  17. [18]N. Arkani-Hamed, D. Baumann, H. Lee, and G. L. Pimentel, The Cosmological Bootstrap: Inflationary Correlators from Symmetries and Singularities, JHEP 2020 (04), 105, arXiv:1811.00024 [hep-th].
  18. [19]Z. Qin and Z.-Z. Xianyu, Closed-form formulae for inflation correlators, JHEP 2023 (07), 001, arXiv:2301.07047 [hep-th].DOI
  19. [20]Z.-Z. Xianyu and J. Zang, Inflation correlators with multiple massive exchanges, JHEP 2024 (03), 070, arXiv:2309.10849 [hep-th].arxiv.org/abs/2309.10849
  20. [21]S. Aoki, L. Pinol, F. Sano, M. Yamaguchi, and Y. Zhu, Cosmological correlators with double massive exchanges: bootstrap equation and phenomenology, JHEP 2024 (09), 176, arXiv:2404.09547 [hep-th].arxiv.org/abs/2404.09547
  21. [22]N. Belrhali, A. Poisson, and S. Renaux-Petel, Cosmological Correlators in KLF and the Double-Exchange (2026), arXiv:2607.05327 [hep-th].arxiv.org/abs/2607.05327
  22. [23]S. Banik and S. Bera, HyperPrecision: a Mathematica package for high-precision numerical evaluation of multivariate hypergeometric functions, Comput. Phys. Commun. 328, 110328 (2026), arXiv:2605.30216 [hep-ph].DOI
  23. [24]H. An, M. McAneny, A. K. Ridgway, and M. B. Wise, Quasi Single Field Inflation in the non-perturbative regime, JHEP 2018 (06), 105, arXiv:1706.09971 [hep-ph].
  24. [25]D. Werth, L. Pinol, and S. Renaux-Petel, Cosmological Flow of Primordial Correlators, Phys. Rev. Lett. 133, 141002 (2024), arXiv:2302.00655 [hep-th].arxiv.org/abs/2302.00655
  25. [26]J. Chen, Y.-X. Tao, X. Wang, and D. Zhang, MadStree: automated evaluation of tree-level de Sitter correlators (2026), arXiv:2610.28526 [hep-ph].arxiv.org/abs/2610.28526
  26. [27]X. Chen, Y. Wang, and Z.-Z. Xianyu, Schwinger-Keldysh Diagrammatics for Primordial Perturbations, JCAP 2017 (12), 006, arXiv:1703.10166 [hep-th].DOI
  27. [28]A. McCune, Massive de Sitter Correlators as Finite Mellin-Barnes Integrals (2026), arXiv:2610.10921 [hep-th].
  28. [29]H. Liu and Z.-Z. Xianyu, Massive inflationary amplitudes: differential equations and complete solutions for general trees, JHEP 2025 (09), 183, arXiv:2412.07843 [hep-th].arxiv.org/abs/2412.07843
  29. [30]Z.-Z. Xianyu and J. Zang, Massive inflationary amplitudes: new representations and degenerate limits, JHEP 2026 (03), 122, arXiv:2511.08677 [hep-th].arxiv.org/abs/2511.08677
  30. [31]J. Gräfe and D. Werth, All Tree-Level Massive Cosmological Correlators via Spectral Gluing (2026), arXiv:2607.18223 [hep-th].arxiv.org/abs/2607.18223
  31. [32]J. Huenupi, C. Muñoz, G. A. Palma, and S. Sypsas, Pushing the Primordial Frontier: Cosmological Collider Signatures at Strong Mixing (2026), arXiv:2607.14529 [astro-ph.CO].
  32. [33]L. Pinol, Exact bispectra in strongly mixed multifield inflation (2026), arXiv:2609.38167 [hep-th].arxiv.org/abs/2609.38167
  33. [34]X. Wang, Y. Wang, and Y. Zhao, Cosmological Collider Signals at Strong Mixing (2026), arXiv:2607.14891 [hep-th].arxiv.org/abs/2607.14891
  34. [35]N. Belrhali, A. Poisson, and S. Renaux-Petel, Analytical Cosmological Collider at Strong Mixing in Laplace Space (2026), arXiv:2608.23243 [hep-th].arxiv.org/abs/2608.23243
  35. [36]X. Chen and Z.-Z. Xianyu, Triple-Massive-Exchange Bispectrum of Quasi-Single-Field Inflation (2026), arXiv:2611.15660 [hep-th].arxiv.org/abs/2611.15660
  36. [37]O. H. E. Philcox, The Limitations of Cosmological Collider Analyses (2026), arXiv:2607.12010 [astro-ph.CO].arxiv.org/abs/2607.12010
  37. [38]A. B. Goncharov, Multiple polylogarithms, cyclotomy and modular complexes, Math. Res. Lett. 5, 497 (1998), arXiv:1105.2076 [math.AG].
  38. [39]O. H. E. Philcox, Searching for inflationary physics with the CMB spectrum. III. Constraints from Planck, Phys. Rev. D 111, 123534 (2025), arXiv:2502.06931 [astro-ph.CO].arxiv.org/abs/2502.06931
  39. [40]C. Cheung, P. Creminelli, A. L. Fitzpatrick, J. Kaplan, and L. Senatore, The Effective Field Theory of Inflation, JHEP 2008 (03), 014, arXiv:0709.0293 [hep-th].DOI
  40. [41]D. Werth, L. Pinol, and S. Renaux-Petel, CosmoFlow: Python Package for Cosmological Correlators, Class. Quant. Grav. 41, 175015 (2024), arXiv:2402.03693 [astro-ph.CO].DOI
  41. [42]T. S. Bunch and P. C. W. Davies, Quantum Field Theory in de Sitter Space: Renormalization by Point Splitting, Proc. Roy. Soc. Lond. A 360, 117 (1978).DOI
  42. [43]D. Babich, P. Creminelli, and M. Zaldarriaga, The Shape of non-Gaussianities, JCAP 2004 (08), 009, arXiv:astro-ph/0405356.DOI
  43. [44]S. Weinberg, Quantum contributions to cosmological correlations, Phys. Rev. D 72, 043514 (2005), arXiv:hep-th/0506236.DOI
  44. [45]L. Di Pietro, V. Gorbenko, and S. Komatsu, Analyticity and unitarity for cosmological correlators, JHEP 2022 [46] X. Tong, Y. Wang, and Y. Zhu, Cutting rule for cosmological collider signals: a bulk evolution perspective, JHEP 2022 (03), 181, arXiv:2112.03448 [hep-th].
  45. [47]NIST Digital Library of Mathematical Functions (2026), accessed October 2026.
  46. [48]A. Katsevich and Z. Sun, Cosmological correlators from the Euclidean sphere (2026), arXiv:2610.28516 [hep-th].arxiv.org/abs/2610.28516
  47. [49]J. Gräfe, M. Nowinski, I. Sachs, and S. P. Venna, Cosmological Dressing Rules for Massive Scalars (2026), arXiv:2611.15870 [hep-th].arxiv.org/abs/2611.15870
  48. [50]V. A. Smirnov, Analytic Tools for Feynman Integrals, Springer Tracts Mod. Phys., Vol. 250 (Springer, 2012).
  49. [51]M. Czakon, Automated analytic continuation of Mellin-Barnes integrals, Comput. Phys. Commun. 175, 559 (2006), arXiv:hep-ph/0511200.DOI
  50. [52]A. V. Smirnov and V. A. Smirnov, On the Resolution of Singularities of Multiple Mellin-Barnes Integrals, Eur. Phys. J. C 62, 445 (2009), arXiv:0901.0386 [hep-ph].DOI
  51. [53]J. B. Tausk, Non-planar massless two-loop Feynman diagrams with four on-shell legs, Phys. Lett. B 469, 425 (1999), arXiv:hep-ph/9909506.DOI
  52. [54]C. Anastasiou and A. Daleo, Numerical evaluation of loop integrals, JHEP 2006 (10), 031, arXiv:hep-ph/0511176.DOI
  53. [55]T. Ahmed, S. Banik, S. M. Hasan, and A. Rapakouliaus, MBAnalytica: Analytic Evaluation of Mellin–Barnes Integrals (2026), in preparation.
  54. [56]T. Ahmed, S. M. Hasan, and A. Rapakouliaus, Phase-space integrals through Mellin-Barnes representation, Phys. Rev. D 112, L051903 (2025), arXiv:2410.18886 [hep-ph].arxiv.org/abs/2410.18886
  55. [57]T. Ahmed, S. M. Hasan, and A. Rapakouliaus, Angular phase-space integrals with four denominators through Mellin-Barnes, JHEP 2025 (11), 152, arXiv:2508.15952 [hep-ph].arxiv.org/abs/2508.15952
  56. [58]B. Ananthanarayan, S. Banik, S. Friot, and S. Ghosh, Multiple Series Representations of N-fold Mellin-Barnes Integrals, Phys. Rev. Lett. 127, 151601 (2021), arXiv:2012.15108 [hep-th].
  57. [59]S. Banik and S. Friot, Multiple Mellin-Barnes integrals with straight contours, Phys. Rev. D 107, 016007 (2023), arXiv:2212.11839 [hep-ph].DOI
  58. [60]S. Banik and S. Friot, Multiple Mellin-Barnes integrals and triangulations of point configurations, Phys. Rev. D 110, 036002 (2024), arXiv:2309.00409 [hep-th].arxiv.org/abs/2309.00409
  59. [61]S. Banik and S. Friot, Multiple Mellin-Barnes integrals with polygamma functions, Phys. Rev. D 114, 056004 (2026), arXiv:2512.19803 [hep-ph].arxiv.org/abs/2512.19803
  60. [62]I. Dubovyk, J. Gluza, and G. Somogyi, Mellin-Barnes Integrals: A Primer on Particle Physics Applications, Lect. Notes Phys. 1008, pp. (2022), arXiv:2211.13733 [hep-ph].
  61. [63]J. Vollinga and S. Weinzierl, Numerical evaluation of multiple polylogarithms, Comput. Phys. Commun. 167, 177 (2005), arXiv:hep-ph/0410259.DOI
  62. [64]N. Arkani-Hamed, P. Benincasa, and A. Postnikov, Cosmological Polytopes and the Wavefunction of the Universe (2017), arXiv:1709.02813 [hep-th].
  63. [65]All closed-form expressions are provided as electronically readable ancillary files with the arXiv submission.
  64. [66]J. Henn, J. Mei, and Q. Yang, A compact analytic formula for the one-loop three-point cosmological correlator (2026), arXiv:2608.17877 [hep-th].arxiv.org/abs/2608.17877
  65. [67]A. B. Goncharov, M. Spradlin, C. Vergu, and A. Volovich, Classical Polylogarithms for Amplitudes andDOI
  66. [68]N. Arkani-Hamed, D. Baumann, A. Hillman, A. Joyce, H. Lee, and G. L. Pimentel, Kinematic Flow and the Emergence of Time, Phys. Rev. Lett. 135, 031602 (2025), arXiv:2312.05300 [hep-th].
  67. [69]N. Arkani-Hamed, D. Baumann, A. Hillman, A. Joyce, H. Lee, and G. L. Pimentel, Differential equations for cosmological correlators, JHEP 2025 (09), 009, arXiv:2312.05303 [hep-th].
  68. [70]D. Baumann, A. Joyce, H. Lee, and K. Salehi Vaziri, Differential Equations for Massive Correlators (2026), arXiv:2604.08658 [hep-th].arxiv.org/abs/2604.08658
  69. [71]E. Komatsu and D. N. Spergel, Acoustic signatures in the primary microwave background bispectrum, Phys. Rev. D 63, 063002 (2001), arXiv:astro-ph/0005036.DOI
  70. [72]D. H. Bailey and D. J. Broadhurst, Parallel integer relation detection: Techniques and applications, Math. Comput. 70, 1719 (2001), arXiv:math/9905048.arxiv.org/abs/math/9905048
  71. [73]D.-G. Wang, G. L. Pimentel, and A. Achúcarro, Bootstrapping multi-field inflation: non-Gaussianities from light scalars revisited, JCAP 2023 (05), 043, arXiv:2212.14035 [astro-ph.CO].DOI
  72. [74]W. Sohn, J. R. Fergusson, and E. P. S. Shellard, High-resolution CMB bispectrum estimator with flexible modal bases, Phys. Rev. D 108, 063504 (2023), arXiv:2305.14646 [astro-ph.CO].DOI
  73. [75]E. Sefusatti, J. R. Fergusson, X. Chen, and E. P. S. Shellard, Effects and Detectability of Quasi-Single Field Inflation in the Large-Scale Structure and Cosmic Microwave Background, JCAP 2012 (08), 033, arXiv:1204.6318 [astro-ph.CO].DOI
  74. [76]Y. Akrami et al. (Planck), Planck 2018 results. IX. Constraints on primordial non-Gaussianity, Astron. Astrophys. 641, A9 (2020), arXiv:1905.05697 [astro-ph.CO].arxiv.org/abs/1905.05697
  75. [77]K. N. Abazajian et al. (CMB-S4), CMB-S4 Science Book, First Edition (2016), arXiv:1610.02743 [astro-ph.CO].arxiv.org/abs/1610.02743
  76. [78]N. Dalal, O. Dore, D. Huterer, and A. Shirokov, The imprints of primordial non-gaussianities on large-scale structure: scale dependent bias and abundance of virialized objects, Phys. Rev. D 77, 123514 (2008), arXiv:0710.4560 [astro-ph].DOI
  77. [79]J. Gleyzes, R. de Putter, D. Green, and O. Doré, Biasing and the search for primordial non-Gaussianity beyond the local type, JCAP 2017 (04), 002, arXiv:1612.06366 [astro-ph.CO].DOI
  78. [80]S. Goldstein, O. H. E. Philcox, J. C. Hill, and L. Hui, Intermediate mass-range particles from small scales: Non-perturbative techniques for cosmological collider physics from large-scale structure surveys, Phys. Rev. D 110, 083516 (2024), arXiv:2407.08731 [astro-ph.CO].arxiv.org/abs/2407.08731
  79. [81]X. Chen, W. Z. Chua, Y. Guo, Y. Wang, Z.-Z. Xianyu, and T. Xie, Quantum Standard Clocks in the Primordial Trispectrum, JCAP 2018 (05), 049, arXiv:1803.04412 [hep-th].DOI
  80. [82]D. Baumann, S. Ferraro, D. Green, and K. M. Smith, Stochastic Bias from Non-Gaussian Initial Conditions, JCAP 2013 (05), 001, arXiv:1209.2173 [astro-ph.CO].DOI

Paper details

vlnc:W_eehqx8en
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents