Introduction
Loop quantum cosmology (LQC) replaces the classical big-bang singularity by a quantum bounce and provides a setting in which cosmological perturbations can be propagated through a Planck-curvature epoch [1 , 2 ] . Two widely used frameworks for cosmological perturbations are the dressed-metric and hybrid approaches. In both cases the homogeneous background is loop quantized, while the inhomogeneous perturbations are quantized in a Fock representation. In the dressed-metric approach, perturbations propagate as quantum fields on an effective geometry dressed by the quantum background [3 , 4 , 5 ] , whereas in the hybrid approach the truncated background-plus-perturbation system is quantized as a constrained system, combining polymeric quantization of the homogeneous sector with Fock quantization of the perturbations [6 , 7 , 8 ] . Although the two formulations are related at the classical level, their effective perturbation theories need not coincide after the background structures entering the perturbative Hamiltonian are regularized [9 , 10 ] .
Modified loop quantum cosmologies arise from regularization ambiguities in the gravitational Hamiltonian constraint. In standard LQC the Euclidean and Lorentzian contributions are first related classically in spatially flat FLRW spacetime and the resulting combined expression is then quantized. An alternative route is to quantize these two contributions separately, more closely following their treatment in full loop quantum gravity. This possibility was first investigated by Yang, Ding, and Ma [11 ] , leading to the two quantizations now known as mLQC-I and mLQC-II. The quantum dynamics of mLQC-I in the improved μ ˉ \bar{\mu} μ ˉ scheme was subsequently developed in Ref. [12 ] . A related effective Hamiltonian also emerged from a top-down construction based on full loop quantum gravity [13 ] , and the improved μ ˉ \bar{\mu} μ ˉ dynamics was later recovered when graph-changing effects were incorporated [14 ] . The effective background dynamics of mLQC-I and mLQC-II were analyzed systematically in Refs. [15 , 16 ] ; see also Ref. [17 ] for a review.
A particularly important feature of mLQC-I is the intrinsic asymmetry of its quantum bounce. The expanding branch approaches classical general relativity at low curvature, whereas the remote contracting phase approaches a quantum-geometrical de Sitter spacetime {1 }. Slow-roll inflation arises generically for kinetic-energy-dominated initial data at the bounce [20 ] , and the background evolution exhibits universal features for a broad class of inflationary potentials [21 ] . This asymmetric pre-bounce structure makes mLQC-I especially interesting for studying possible quantum-gravitational imprints in primordial perturbations.
Cosmological perturbations in mLQC-I have consequently been investigated in both the dressed-metric and hybrid approaches. Within the dressed-metric framework, the primordial power spectrum was studied in Ref. [22 ] , followed by a systematic comparison of the scalar and tensor spectra of LQC, mLQC-I, and mLQC-II [23 ] . Within the hybrid framework, the perturbative construction was extended to mLQC-I in Ref. [24 ] , while the corresponding scalar and tensor effective masses and their relation to the dressed-metric description were analyzed in Ref. [25 ] . The primordial scalar spectra of the three loop cosmologies were compared within the hybrid framework in Ref. [26 ] . The pre-bounce perturbation dynamics and the selection of stable initial states were investigated further in Ref. [27 ] . More recently, high-accuracy mode functions, ultraviolet regularity, and particle creation in the hybrid approach to mLQC-I have been studied in Ref. [28 ] .
Beyond the power spectrum, primordial non-Gaussianity provides an independent and potentially more sensitive probe of pre-inflationary physics. In single-field slow-roll inflation the primordial bispectrum is strongly constrained by the approximate de Sitter symmetries and is normally small [29 ] . Quantum-bounce dynamics can lead to qualitatively different behavior. Primordial non-Gaussianity in standard LQC and related bouncing scenarios has been investigated in Refs. [30 , 31 , 32 , 33 , 34 , 35 , 36 ] . These studies showed that the bounce can generate strongly scale-dependent and oscillatory non-Gaussianity, particularly on scales that are sensitive to the pre-inflationary quantum-gravity regime. To our knowledge, however, a corresponding nonlinear bispectrum calculation has not previously been carried out for either mLQC-I or mLQC-II.
The observational interpretation of such a primordial bispectrum requires some care. A large pointwise value of f N L f_{\mathrm{NL}} f NL need not translate into a large CMB bispectrum when the primordial signal oscillates on momentum scales much shorter than those resolved by the radiation-transfer kernels. For standard LQC, Roshna and Sreenath showed explicitly that rapid primordial oscillations can be strongly averaged in the reduced CMB bispectrum [37 ] . On the other hand, smoother bouncing bispectrum templates can survive the transfer projection much more efficiently and may be strongly constrained by CMB data [38 ] . It is therefore important to determine not only the magnitude but also the momentum-space phase and triangle dependence of the primordial bispectrum.
A logically distinct issue is whether the perturbative expansion itself remains under control when the primordial three-point function becomes large. Agullo, Bolliet and Sreenath (ABS) addressed this question in standard LQC by computing the leading nonlinear correction to the curvature power spectrum and comparing it with the leading-order result [31 ] . Their analysis demonstrated that a large primordial bispectrum does not by itself imply a breakdown of the correlation-function expansion, because the correction to the two-point function contains additional powers of the small scalar power spectrum. This motivates the diagnostic
Ξ ( k ) ≡ ∣ Δ P R ( k ) P R ( 0 ) ( k ) ∣ , (1) \Xi(k) \equiv\left|\frac{\Delta\mathcal{P}_{\mathcal{R}}(k)}{\mathcal{P}_{\mathcal{R}}^{(0)}(k)}\right|, \tag*{(1)} Ξ ( k ) ≡ P R ( 0 ) ( k ) Δ P R ( k ) , (1)
with Ξ ≪ 1 \Xi\ll1 Ξ ≪ 1 testing the smallness of the retained nonlinear-conversion correction on the stated domain.
The purpose of the present work is to extend both aspects of this nonlinear analysis to mLQC-I within the hybrid approach. First, we construct the cubic interaction Hamiltonian in the same hybrid canonical variables used in the quadratic theory and derive the exact finite-momentum tree-level bispectrum. The ordering of operations is important: the classical quadratic and cubic Hamiltonians are first written in a common canonical representation without using the classical Friedmann constraint, the mLQC-I regularization is then implemented, and only afterward is the effective background constraint used. Second, we evaluate the ABS correlation-function diagnostic for two physically distinct initial-state prescriptions: a Gaussian state specified directly at the bounce and the de Sitter vacuum selected in the remote contracting phase. We also retain the complete second-order relation between the spatially flat-gauge field perturbation and the comoving curvature perturbation and construct its minimal mLQC-I effective continuation, providing the basis for future nonlinear calculations in the contracting phase and near the bounce2 .
Our results show that these two initial-state prescriptions lead to very different degrees of cancellation sensitivity in the retained correlation-function diagnostic. For the bounce-defined state, the finite-window absolute-integrand estimate is below 6 × 10 − 6 6 \times10^{-6} 6 × 1 0 − 6 at the five sampled momenta. For the remote de Sitter state, the signed correction is more oscillatory and cancellation dominated, decreasing strongly as the external scale approaches the bounce-curvature scale. At the same time, the primordial bispectrum can be very large pointwise. For the bounce-defined state the phase root-mean-square (RMS) equilateral f N L f_{\mathrm{NL}} f NL is typically of order 10 3 10^{3} 1 0 3 –10 4 10^{4} 1 0 4 over the displayed range, while the remote prescription can give much larger values.
One of our central numerical results is that the rapidly varying part of the bispectrum is approximately organized by the total momentum k t = k 1 + k 2 + k 3 k_{\mathrm{t}}=k_{1}+k_{2}+k_{3} k t = k 1 + k 2 + k 3 , with a nearly universal oscillation period Δ k t ≃ 5.7 × 10 − 4 m P \Delta k_{\mathrm{t}}\simeq5.7\times10^{-4}m_{\mathrm{P}} Δ k t ≃ 5.7 × 1 0 − 4 m P over the triangle configurations studied, whereas the momentum-triangle envelope and relative phase vary much more slowly. This separation motivates a general numerical template for the primordial non-Gaussianity. For the illustrative bounce-to-present mapping used in our projection analysis, the corresponding oscillations are much faster than the CMB transfer scales, and our thin-shell and acoustic-transfer diagnostics indicate substantial phase averaging.
The remainder of the paper is organized as follows. In Sec. II we write the classical quadratic and cubic Hamiltonians in a common set of hybrid canonical variables and present the complete nonlinear relation between the field perturbation and the co-moving curvature perturbation R \mathcal{R} R . In Sec. III we implement the mLQC-I effective quantization, construct the cubic interaction Hamiltonian, derive the minimal effective continuation of the nonlinear curvature perturbation, and examine formal symmetry and the classical limit. Sec. IV derives the exact finite-momentum cubic vertex, tree-level bispectrum, and the retained nonlinear-conversion diagnostic. In Sec. V we specify the bounce-defined and remote-contracting initial states and evaluate their perturbativity properties. Sec. VI studies the primordial power spectrum and bispectrum, the rapid k t k_t k t oscillation, the general momentum-triangle template, and CMB-projection diagnostics. Our conclusions are summarized in Sec. VII. Four appendices provide useful algebraic relations, the canonical completion, interpolation details, and independent numerical validation.
Throughout this paper we set ℏ = c = 1 \hbar=c=1 ℏ = c = 1 and use κ = 8 π G \kappa=8\pi G κ = 8 π G . We denote the unreduced Planck mass and Planck time by m P ≡ G − 1 / 2 m_{\mathrm{P}}\equiv G^{-1/2} m P ≡ G − 1/2 and t P ≡ G 1 / 2 t_{\mathrm{P}}\equiv G^{1/2} t P ≡ G 1/2 , respectively.
Classical Hamiltonians in the hybrid canonical variables
We write
v = a 3 , Ω ≡ Ω c a n = v b = γ a 3 H c l , (2) v=a^3,\qquad\Omega\equiv\Omega_{\mathrm{can}}=vb=\gamma a^3H_{\mathrm{cl}}, \tag*{(2)} v = a 3 , Ω ≡ Ω can = v b = γ a 3 H cl , (2)
and use the fiducial Euclidean convention
∂ 2 ≡ δ i j ∂ i ∂ j , ( ∂ f ) 2 ≡ δ i j ( ∂ i f ) ( ∂ j f ) , (3) \partial^2\equiv\delta^{ij}\partial_i\partial_j,\qquad(\partial f)^2\equiv\delta^{ij}(\partial_i f)(\partial_j f), \tag*{(3)} ∂ 2 ≡ δ ij ∂ i ∂ j , ( ∂ f ) 2 ≡ δ ij ( ∂ i f ) ( ∂ j f ) , (3)
for spatial derivatives in the transformed hybrid Hamiltonian. Here H c l = b / γ H_{\mathrm{cl}}=b/\gamma H cl = b / γ is the classical canonical relation; it is not the effective mLQC-I rate. Throughout the classical phase-space substitutions, Ω \Omega Ω means Ω c a n \Omega_{\mathrm{can}} Ω can . The rate-based comparison variable Ω H = γ v H e f f \Omega_H=\gamma vH_{\mathrm{eff}} Ω H = γ v H eff will be used explicitly only in Sec. III G.
Quadratic Hamiltonian
The quadratic Hamiltonian in the Mukhanov–Sasaki (MS) canonical pair ( ν , π ν ) (\nu,\pi_\nu) ( ν , π ν ) can be written as [6 , 7 , 8 ]
H c l ( 2 ) = N 2 a ∫ d 3 x [ π ν 2 + ( ∂ ν ) 2 + m c l 2 ν 2 ] . (4) H_{\mathrm{cl}}^{(2)}=\frac{N}{2a}\int d^3x\left[\pi_\nu^2+(\partial\nu)^2+m_{\mathrm{cl}}^2\nu^2\right]. \tag*{(4)} H cl ( 2 ) = 2 a N ∫ d 3 x [ π ν 2 + ( ∂ ν ) 2 + m cl 2 ν 2 ] . (4)
We use the off-shell classical mass in the canonical variables ( a , π a ; ϕ , p ϕ ) (a,\pi_a;\phi,p_\phi) ( a , π a ; ϕ , p ϕ ) ,
m c l 2 = − 27 p ϕ 4 2 π 2 a 6 + 5 κ p ϕ 2 2 a 4 + 9 p ϕ 2 V π a 2 − 12 a p ϕ V , ϕ π a − κ 2 π a 2 72 a 2 + a 2 V , ϕ ϕ − κ 2 a 2 V . (5) m_{\mathrm{cl}}^2=-\frac{27p_\phi^4}{2\pi^2a^6}+\frac{5\kappa p_\phi^2}{2a^4}+\frac{9p_\phi^2V}{\pi_a^2}-\frac{12ap_\phi V_{,\phi}}{\pi_a}-\frac{\kappa^2\pi_a^2}{72a^2}+a^2V_{,\phi\phi}-\frac{\kappa}{2}a^2V. \tag*{(5)} m cl 2 = − 2 π 2 a 6 27 p ϕ 4 + 2 a 4 5 κ p ϕ 2 + π a 2 9 p ϕ 2 V − π a 12 a p ϕ V , ϕ − 72 a 2 κ 2 π a 2 + a 2 V , ϕϕ − 2 κ a 2 V . (5)
No use of the classical background Hamiltonian constraint is made in (5) . This is the form that will be carried into the effective construction.
Using the classical relation
π a = − 6 Ω κ γ a , (6) \pi_a=-\frac{6\Omega}{\kappa\gamma a}, \tag*{(6)} π a = − κγ a 6Ω , (6)
(5) can equivalently be written as
m c l 2 = − 3 κ 2 γ 2 p ϕ 4 8 v 4 / 3 Ω 2 + 5 κ p ϕ 2 2 v 4 / 3 + κ 2 γ 2 v 2 / 3 p ϕ 2 V 4 Ω 2 + 2 κ γ v 2 / 3 p ϕ Ω V , ϕ − Ω 2 2 γ 2 v 4 / 3 + v 2 / 3 V , ϕ ϕ − κ 2 v 2 / 3 V . (7) m_{\mathrm{cl}}^2=-\frac{3\kappa^2\gamma^2p_\phi^4}{8v^{4/3}\Omega^2}+\frac{5\kappa p_\phi^2}{2v^{4/3}}+\frac{\kappa^2\gamma^2v^{2/3}p_\phi^2V}{4\Omega^2}+\frac{2\kappa\gamma v^{2/3}p_\phi}{\Omega}V_{,\phi}-\frac{\Omega^2}{2\gamma^2v^{4/3}}+v^{2/3}V_{,\phi\phi}-\frac{\kappa}{2}v^{2/3}V. \tag*{(7)} m cl 2 = − 8 v 4/3 Ω 2 3 κ 2 γ 2 p ϕ 4 + 2 v 4/3 5 κ p ϕ 2 + 4 Ω 2 κ 2 γ 2 v 2/3 p ϕ 2 V + Ω 2 κγ v 2/3 p ϕ V , ϕ − 2 γ 2 v 4/3 Ω 2 + v 2/3 V , ϕϕ − 2 κ v 2/3 V . (7)
Classical cubic Hamiltonian in the spatially flat gauge
A convenient off-shell third-order Hamiltonian for scalar perturbations in the spatially flat gauge is given in Ref. [31 ] . The background Friedmann constraint is not used to simplify either the quadratic or cubic coefficients at this stage, which is essential for applying the mLQC-I regularization consistently. We also retain explicitly the momentum redefinition made before the displayed cubic Hamiltonian in that reference, writing
δ p ˉ ϕ = δ p ϕ o l d − 3 p ϕ 2 a π a δ ϕ . (8) \delta\bar p_\phi=\delta p_\phi^{\mathrm{old}}-\frac{3p_\phi^2}{a\pi_a}\delta\phi. \tag*{(8)} δ p ˉ ϕ = δ p ϕ old − a π a 3 p ϕ 2 δ ϕ . (8)
Thus, in the expression below, the momentum perturbation is understood as δ p ˉ ϕ \delta\bar p_\phi δ p ˉ ϕ . Before transforming to the hybrid variables, the cubic Hamiltonian reads
H c l ( 3 ) = N ∫ d 3 x { ( 9 κ p ϕ 3 4 a 4 π a − 27 p ϕ 5 2 a 6 π a 3 − 3 a 2 p ϕ V , ϕ ϕ 2 π a + a 3 V , ϕ ϕ ϕ 6 ) ( δ ϕ ) 3 − 3 p ϕ 2 a 4 π a ( δ p ˉ ϕ ) 2 δ ϕ − 9 p ϕ 3 a 5 π a 3 δ p ˉ ϕ ( δ ϕ ) 2 − 3 a 2 p ϕ 2 π a δ ϕ ( ∂ δ ϕ ) 2 + 3 p ϕ 2 N a π a ( δ ϕ ) 2 ∂ 2 χ + 3 a 2 p ϕ 2 N 2 κ π a δ ϕ ( ∂ 2 χ ) 2 + 3 p ϕ 2 N a π a δ ϕ ∂ i χ ∂ i δ ϕ + 1 N δ p ˉ ϕ ∂ i δ ϕ ∂ i χ − 3 a 2 p ϕ 2 N 2 κ π a δ ϕ ∂ i ∂ j χ ∂ i ∂ j χ } . (9) H_{\mathrm{cl}}^{(3)}=N\int d^3x\left\{\left(\frac{9\kappa p_\phi^3}{4a^4\pi_a}-\frac{27p_\phi^5}{2a^6\pi_a^3}-\frac{3a^2p_\phi V_{,\phi\phi}}{2\pi_a}+\frac{a^3V_{,\phi\phi\phi}}{6}\right)(\delta\phi)^3-\frac{3p_\phi}{2a^4\pi_a}(\delta\bar p_\phi)^2\delta\phi-\frac{9p_\phi^3}{a^5\pi_a^3}\delta\bar p_\phi(\delta\phi)^2-\frac{3a^2p_\phi}{2\pi_a}\delta\phi(\partial\delta\phi)^2+\frac{3p_\phi^2}{Na\pi_a}(\delta\phi)^2\partial^2\chi+\frac{3a^2p_\phi}{2N^2\kappa\pi_a}\delta\phi(\partial^2\chi)^2+\frac{3p_\phi^2}{Na\pi_a}\delta\phi\,\partial_i\chi\,\partial^i\delta\phi+\frac{1}{N}\delta\bar p_\phi\,\partial_i\delta\phi\,\partial^i\chi-\frac{3a^2p_\phi}{2N^2\kappa\pi_a}\delta\phi\,\partial_i\partial_j\chi\,\partial^i\partial^j\chi\right\}. \tag*{(9)} H cl ( 3 ) = N ∫ d 3 x { ( 4 a 4 π a 9 κ p ϕ 3 − 2 a 6 π a 3 27 p ϕ 5 − 2 π a 3 a 2 p ϕ V , ϕϕ + 6 a 3 V , ϕϕϕ ) ( δ ϕ ) 3 − 2 a 4 π a 3 p ϕ ( δ p ˉ ϕ ) 2 δ ϕ − a 5 π a 3 9 p ϕ 3 δ p ˉ ϕ ( δ ϕ ) 2 − 2 π a 3 a 2 p ϕ δ ϕ ( ∂ δ ϕ ) 2 + N a π a 3 p ϕ 2 ( δ ϕ ) 2 ∂ 2 χ + 2 N 2 κ π a 3 a 2 p ϕ δ ϕ ( ∂ 2 χ ) 2 + N a π a 3 p ϕ 2 δ ϕ ∂ i χ ∂ i δ ϕ + N 1 δ p ˉ ϕ ∂ i δ ϕ ∂ i χ − 2 N 2 κ π a 3 a 2 p ϕ δ ϕ ∂ i ∂ j χ ∂ i ∂ j χ } . (9)
In Eq. (9) the derivatives and index contractions are written with the physical FLRW spatial metric. They will be converted below to the fiducial Euclidean convention used for the hybrid Hamiltonian.
The scalar shift potential χ \chi χ follows from the first-order constraints. After restoring the momentum variable in Eq. (8) , its Fourier mode is
χ ~ k = 3 N κ k 2 a [ p ϕ 2 − a 5 V , ϕ κ π a − 3 p ϕ 3 κ a 2 π a 2 ] δ ϕ ~ k − p ϕ κ a π a δ p ˉ ~ ϕ k . (10) \widetilde{\chi}_{\mathbf{k}}=\frac{3N\kappa}{k^2a}\left[\frac{p_\phi}{2}-\frac{a^5V_{,\phi}}{\kappa\pi_a}-\frac{3p_\phi^3}{\kappa a^2\pi_a^2}\right]\delta\widetilde{\phi}_{\mathbf{k}}-\frac{p_\phi}{\kappa a\pi_a}\delta\widetilde{\bar p}_{\phi\mathbf{k}}. \tag*{(10)} χ k = k 2 a 3 N κ [ 2 p ϕ − κ π a a 5 V , ϕ − κ a 2 π a 2 3 p ϕ 3 ] δ ϕ k − κa π a p ϕ δ p ˉ ϕ k . (10)
The MS canonical pair used in the hybrid approach is related to the variables above by
ν = a δ ϕ , (11) \nu=a\,\delta\phi, \tag*{(11)} ν = a δ ϕ , (11)
and
π ν = δ p ˉ ϕ a − κ π a 6 a ν . (12) \pi_\nu=\frac{\delta\bar p_\phi}{a}-\frac{\kappa\pi_a}{6a}\nu. \tag*{(12)} π ν = a δ p ˉ ϕ − 6 a κ π a ν . (12)
Equivalently,
δ ϕ = ν a , δ p ˉ ϕ = a ( π ν + κ π a 6 a ν ) . (13) \delta\phi=\frac{\nu}{a},\qquad\delta\bar p_\phi=a\left(\pi_\nu+\frac{\kappa\pi_a}{6a}\nu\right). \tag*{(13)} δ ϕ = a ν , δ p ˉ ϕ = a ( π ν + 6 a κ π a ν ) . (13)
Using Eq. (6) , we define the convenient shorthand
P ν ≡ π ν − Ω γ v 2 / 3 ν , (14) \mathcal{P}_\nu\equiv\pi_\nu-\frac{\Omega}{\gamma v^{2/3}}\nu, \tag*{(14)} P ν ≡ π ν − γ v 2/3 Ω ν , (14)
so that
δ p ˉ ϕ = v 1 / 3 P ν . (15) \delta\bar p_\phi=v^{1/3}\mathcal{P}_\nu. \tag*{(15)} δ p ˉ ϕ = v 1/3 P ν . (15)
The canonical transformation on the full background-plus-perturbation phase space requires a little care. In particular, the momentum redefinition in Eq. (8) and the transformation in Eqs. (11) and (12) should be regarded as a single transformation from the original perturbative pair ( δ ϕ , δ p ϕ o l d ) (\delta\phi,\delta p_\phi^{\mathrm{old}}) ( δ ϕ , δ p ϕ old ) to ( ν , π ν ) (\nu,\pi_\nu) ( ν , π ν ) . Explicitly, we have
ν = a δ ϕ , π ν = δ p ϕ o l d a − 3 p ϕ 2 a 2 π a δ ϕ − κ π a 6 δ ϕ , (16) \nu=a\,\delta\phi,\qquad\pi_\nu=\frac{\delta p_\phi^{\mathrm{old}}}{a}-\frac{3p_\phi^2}{a^2\pi_a}\delta\phi-\frac{\kappa\pi_a}{6}\delta\phi, \tag*{(16)} ν = a δ ϕ , π ν = a δ p ϕ old − a 2 π a 3 p ϕ 2 δ ϕ − 6 κ π a δ ϕ , (16)
or equivalently,
δ ϕ = ν a , δ p ϕ o l d = a π ν + ( κ π a 6 + 3 p ϕ 2 a 2 π a ) ν . (17) \delta\phi=\frac{\nu}{a},\qquad\delta p_\phi^{\mathrm{old}}=a\pi_\nu+\left(\frac{\kappa\pi_a}{6}+\frac{3p_\phi^2}{a^2\pi_a}\right)\nu. \tag*{(17)} δ ϕ = a ν , δ p ϕ old = a π ν + ( 6 κ π a + a 2 π a 3 p ϕ 2 ) ν . (17)
Because the coefficients in this transformation depend on the homogeneous variables, completing it canonically on the full phase space induces quadratic perturbative redefinitions of the homogeneous variables. As shown in Appendix B, these shifts modify the background Hamiltonian H c l ( 0 ) H_{\mathrm{cl}}^{(0)} H cl ( 0 ) at second perturbative order and therefore contribute to the quadratic hybrid Hamiltonian. Their insertion into H c l ( 2 ) H_{\mathrm{cl}}^{(2)} H cl ( 2 ) , however, starts at fourth order, whereas their insertion into H c l ( 3 ) H_{\mathrm{cl}}^{(3)} H cl ( 3 ) starts at fifth order. Furthermore, after gauge fixing and imposition of the first-order constraints the reduced perturbative Hamiltonian starts at quadratic order, H p e r t = H ( 2 ) + H ( 3 ) + ⋯ H_{\mathrm{pert}}=H^{(2)}+H^{(3)}+\cdots H pert = H ( 2 ) + H ( 3 ) + ⋯ , so there is no independent first-order contribution that could combine with the quadratic background shifts to generate an additional cubic term. Consequently, within the minimal extension in which ν = a δ ϕ \nu=a\,\delta\phi ν = a δ ϕ remains linear, the full canonical completion generates no additional contribution to the cubic Hamiltonian in Eq. (19) , given below.
We also define
ϑ ≡ χ N . (18) \vartheta\equiv\frac{\chi}{N}. \tag*{(18)} ϑ ≡ N χ . (18)
After converting all spatial contractions to the convention of Eq. (3) , the classical cubic Hamiltonian becomes
H c l ( 3 ) = N ∫ d 3 x { A ν 3 + B ν P ν 2 − C ν P ν + B ν ( ∂ ν ) 2 − D [ ν 2 ∂ 2 ϑ + ν ∂ i ϑ ∂ i ν ] + 1 v 2 / 3 P ν ∂ i ν ∂ i ϑ + E ν [ ( ∂ i ∂ j ϑ ) ( ∂ i ∂ j ϑ ) − ( ∂ 2 ϑ ) 2 ] } . (19) H_{\mathrm{cl}}^{(3)}=N\int d^3x\left\{\mathcal{A}\nu^3+\mathcal{B}\nu\mathcal{P}_\nu^2-\mathcal{C}\nu\mathcal{P}_\nu+\mathcal{B}\nu(\partial\nu)^2-\mathcal{D}\left[\nu^2\partial^2\vartheta+\nu\,\partial_i\vartheta\,\partial^i\nu\right]+\frac{1}{v^{2/3}}\mathcal{P}_\nu\,\partial_i\nu\,\partial^i\vartheta+\mathcal{E}\nu\left[(\partial_i\partial_j\vartheta)(\partial^i\partial^j\vartheta)-(\partial^2\vartheta)^2\right]\right\}. \tag*{(19)} H cl ( 3 ) = N ∫ d 3 x { A ν 3 + B ν P ν 2 − C ν P ν + B ν ( ∂ ν ) 2 − D [ ν 2 ∂ 2 ϑ + ν ∂ i ϑ ∂ i ν ] + v 2/3 1 P ν ∂ i ν ∂ i ϑ + E ν [ ( ∂ i ∂ j ϑ ) ( ∂ i ∂ j ϑ ) − ( ∂ 2 ϑ ) 2 ] } . (19)
where
A = − 3 κ 2 v p ϕ 3 8 v 2 / 3 Ω + κ 3 γ 3 p ϕ 5 16 v 2 / 3 Ω 3 + κ γ p ϕ 4 Ω V , ϕ ϕ + 1 6 V , ϕ ϕ ϕ , B = κ γ p ϕ 4 v 2 / 3 Ω , C = κ 2 γ 2 p ϕ 3 4 v 4 / 3 Ω 2 , D = κ γ p ϕ 2 2 v 4 / 3 Ω , E = γ p ϕ 4 v 2 / 3 Ω . (20) \mathcal{A}=-\frac{3\kappa^{2}v p_{\phi}^{3}}{8v^{2/3}\Omega}+\frac{\kappa^{3}\gamma^{3}p_{\phi}^{5}}{16v^{2/3}\Omega^{3}}+\frac{\kappa\gamma p_{\phi}}{4\Omega}V_{,\phi\phi}+\frac{1}{6}V_{,\phi\phi\phi},
\qquad
\mathcal{B}=\frac{\kappa\gamma p_{\phi}}{4v^{2/3}\Omega},
\qquad
\mathcal{C}=\frac{\kappa^{2}\gamma^{2}p_{\phi}^{3}}{4v^{4/3}\Omega^{2}},
\qquad
\mathcal{D}=\frac{\kappa\gamma p_{\phi}^{2}}{2v^{4/3}\Omega},
\qquad
\mathcal{E}=\frac{\gamma p_{\phi}}{4v^{2/3}\Omega}. \tag*{(20)} A = − 8 v 2/3 Ω 3 κ 2 v p ϕ 3 + 16 v 2/3 Ω 3 κ 3 γ 3 p ϕ 5 + 4Ω κγ p ϕ V , ϕϕ + 6 1 V , ϕϕϕ , B = 4 v 2/3 Ω κγ p ϕ , C = 4 v 4/3 Ω 2 κ 2 γ 2 p ϕ 3 , D = 2 v 4/3 Ω κγ p ϕ 2 , E = 4 v 2/3 Ω γ p ϕ . (20)
The corresponding Fourier-space shift variable is
ϑ k = 3 κ k 2 { v − 2 / 3 [ p ϕ 2 + γ v 2 6 Ω V , ϕ − κ γ 2 p ϕ 3 12 Ω 2 ] v k + γ p ϕ 6 Ω v k } . (21) \vartheta_{\mathbf{k}}=\frac{3\kappa}{k^{2}}\left\{v^{-2/3}\left[\frac{p_{\phi}}{2}+\frac{\gamma v^{2}}{6\Omega}V_{,\phi}-\frac{\kappa\gamma^{2}p_{\phi}^{3}}{12\Omega^{2}}\right]v_{\mathbf{k}}+\frac{\gamma p_{\phi}}{6\Omega}v_{\mathbf{k}}\right\}. \tag*{(21)} ϑ k = k 2 3 κ { v − 2/3 [ 2 p ϕ + 6Ω γ v 2 V , ϕ − 12 Ω 2 κ γ 2 p ϕ 3 ] v k + 6Ω γ p ϕ v k } . (21)
Equations (19) –(21) are the classical cubic Hamiltonian written entirely in the same canonical representation used for the hybrid quadratic theory, before any loop regularization is applied.
Nonlinear relation to the comoving curvature perturbation
The nonlinear relation between the spatially flat-gauge field perturbation and the comoving curvature perturbation is a classical gauge transformation. Because the same relation will be needed in a future analysis of the contracting phase and of the bounce itself, we keep the complete second-order expression rather than introducing only its late-time super-Hubble reduction. We first define
z = − 6 p ϕ κ π a = γ a p ϕ Ω = a ϕ ˙ H . (22) z=-\frac{6p_{\phi}}{\kappa\pi_{a}}=\frac{\gamma a p_{\phi}}{\Omega}=\frac{a\dot{\phi}}{H}. \tag*{(22)} z = − κ π a 6 p ϕ = Ω γ a p ϕ = H a ϕ ˙ . (22)
Since v = a δ ϕ v=a\delta\phi v = a δ ϕ , the linear relation is
R ( 1 ) = − v z . (23) \mathcal{R}^{(1)}=-\frac{v}{z}. \tag*{(23)} R ( 1 ) = − z v . (23)
It is also useful to define the coefficient of the local quadratic term,
C R = − 3 2 + 3 V , ϕ a 5 κ p ϕ π a + κ z 2 4 a 2 . (24) C_{\mathcal{R}}=-\frac{3}{2}+\frac{3V_{,\phi}a^{5}}{\kappa p_{\phi}\pi_{a}}+\frac{\kappa z^{2}}{4a^{2}}. \tag*{(24)} C R = − 2 3 + κ p ϕ π a 3 V , ϕ a 5 + 4 a 2 κ z 2 . (24)
Then the complete second-order transformation of Refs. [29 , 31 ] is
R ( x , t ) = − a z δ ϕ + [ − 3 2 + 3 V , ϕ a 5 κ p ϕ π a + κ z 2 4 a 2 ] ( a z δ ϕ ) 2 − 3 a 2 κ π a d d t ( a z δ ϕ ) 2 − 9 a 4 κ 2 π a 2 a 2 z 2 ( ∂ δ ϕ ) 2 + 9 a 4 κ 2 π a 2 a 2 z 2 ∂ − 2 ∂ i ∂ j ( ∂ i δ ϕ ∂ j δ ϕ ) + 3 a 4 κ π a a z ∂ i χ ∂ i δ ϕ − 3 a 4 κ π a a z ∂ − 2 ∂ i ∂ j ( ∂ i χ ∂ j δ ϕ ) . (25) \mathcal{R}(\mathbf{x},t)=-\frac{a}{z}\delta\phi+\left[-\frac{3}{2}+\frac{3V_{,\phi}a^{5}}{\kappa p_{\phi}\pi_{a}}+\frac{\kappa z^{2}}{4a^{2}}\right]\left(\frac{a}{z}\delta\phi\right)^{2}-\frac{3a^{2}}{\kappa\pi_{a}}\frac{d}{dt}\left(\frac{a}{z}\delta\phi\right)^{2}-\frac{9a^{4}}{\kappa^{2}\pi_{a}^{2}}\frac{a^{2}}{z^{2}}(\partial\delta\phi)^{2}
+\frac{9a^{4}}{\kappa^{2}\pi_{a}^{2}}\frac{a^{2}}{z^{2}}\partial^{-2}\partial_{i}\partial_{j}\left(\partial^{i}\delta\phi\,\partial^{j}\delta\phi\right)+\frac{3a^{4}}{\kappa\pi_{a}}\frac{a}{z}\partial_{i}\chi\,\partial^{i}\delta\phi-\frac{3a^{4}}{\kappa\pi_{a}}\frac{a}{z}\partial^{-2}\partial_{i}\partial_{j}\left(\partial^{i}\chi\,\partial^{j}\delta\phi\right). \tag*{(25)} R ( x , t ) = − z a δ ϕ + [ − 2 3 + κ p ϕ π a 3 V , ϕ a 5 + 4 a 2 κ z 2 ] ( z a δ ϕ ) 2 − κ π a 3 a 2 d t d ( z a δ ϕ ) 2 − κ 2 π a 2 9 a 4 z 2 a 2 ( ∂ δ ϕ ) 2 + κ 2 π a 2 9 a 4 z 2 a 2 ∂ − 2 ∂ i ∂ j ( ∂ i δ ϕ ∂ j δ ϕ ) + κ π a 3 a 4 z a ∂ i χ ∂ i δ ϕ − κ π a 3 a 4 z a ∂ − 2 ∂ i ∂ j ( ∂ i χ ∂ j δ ϕ ) . (25)
Here ∂ − 2 \partial^{-2} ∂ − 2 denotes the inverse spatial Laplacian and χ = N ϑ \chi=N\vartheta χ = N ϑ , with ϑ \vartheta ϑ given in Eq. (21) . In Eq. (25) we retain the convention of Ref. [31 ] : spatial indices in the second-order gauge transformation are raised and lowered with the physical FLRW spatial metric q i j = a 2 δ i j q_{ij}=a^{2}\delta_{ij} q ij = a 2 δ ij . This convention should not be confused with the fiducial-Euclidean derivative convention introduced in Eq. (3) for the transformed hybrid Hamiltonian. Equation (25) , rather than its local reduction below, is the relation relevant when the non-Gaussianity is evaluated in the contracting phase or through the bounce: the time-derivative, gradient, inverse-Laplacian, and shift terms cannot then be discarded a priori.
mLQC-I effective quantization
Background effective Hamiltonian
We start from the classical homogeneous Hamiltonian
H c l ( 0 ) = − 3 Ω 2 κ v γ 2 + p ϕ 2 2 v + v V . (26) H_{\mathrm{cl}}^{(0)}=-\frac{3\Omega^{2}}{\kappa v\gamma^{2}}+\frac{p_{\phi}^{2}}{2v}+vV. \tag*{(26)} H cl ( 0 ) = − κ v γ 2 3 Ω 2 + 2 v p ϕ 2 + v V . (26)
and the mLQC-I effective replacement
Ω 2 → Ω I 2 = − v 2 γ 2 λ 2 { sin 2 ( λ b ) − γ 2 + 1 4 γ 2 sin 2 ( 2 λ b ) } . (27) \Omega^{2}\rightarrow\Omega_{\mathrm{I}}^{2}=-\frac{v^{2}\gamma^{2}}{\lambda^{2}}\left\{\sin^{2}(\lambda b)-\frac{\gamma^{2}+1}{4\gamma^{2}}\sin^{2}(2\lambda b)\right\}. \tag*{(27)} Ω 2 → Ω I 2 = − λ 2 v 2 γ 2 { sin 2 ( λb ) − 4 γ 2 γ 2 + 1 sin 2 ( 2 λb ) } . (27)
Thus, we find
H e f f ( 0 ) = − 3 Ω I 2 κ v γ 2 + p ϕ 2 2 v + v V . (28) H_{\mathrm{eff}}^{(0)}=-\frac{3\Omega_{\mathrm{I}}^{2}}{\kappa v\gamma^{2}}+\frac{p_{\phi}^{2}}{2v}+vV. \tag*{(28)} H eff ( 0 ) = − κ v γ 2 3 Ω I 2 + 2 v p ϕ 2 + v V . (28)
It is useful to define
Λ I ≡ v 2 λ sin ( 2 λ b ) . (29) \Lambda_{\mathrm{I}}\equiv\frac{v}{2\lambda}\sin(2\lambda b). \tag*{(29)} Λ I ≡ 2 λ v sin ( 2 λb ) . (29)
The perturbative prescription used in the hybrid mLQC-I construction is
1 Ω ^ 2 ⟶ 1 Ω ^ I 2 , 1 Ω ^ ⟶ Λ ^ I Ω ^ I 2 . (30) \frac{1}{\widehat{\Omega}^{2}} \longrightarrow\frac{1}{\widehat{\Omega}_{\mathrm{I}}^{2}}, \qquad\frac{1}{\widehat{\Omega}} \longrightarrow\frac{\widehat{\Lambda}_{\mathrm{I}}}{\widehat{\Omega}_{\mathrm{I}}^{2}}. \tag*{(30)} Ω 2 1 ⟶ Ω I 2 1 , Ω 1 ⟶ Ω I 2 Λ I . (30)
The first rule is tied to the mLQC-I background regularization, while the second is chosen so that the perturbative operator respects the background superselection structure [26 ] . The same quadratic prescription is used in our particle-creation analysis [28 ] .
Quadratic effective Hamiltonian and consistency check
Applying Eqs. (27) and (30) directly to the off-shell expression (7) gives
m e f f , r a w 2 = − 3 κ 2 γ 2 v 2 / 3 p ϕ 4 8 v 4 / 3 Ω I 2 + 5 κ p ϕ 2 2 v 4 / 3 + κ 2 γ 2 v 2 / 3 p ϕ 2 4 Ω I 2 + 2 κ γ v 2 / 3 p ϕ Λ I Ω I 2 V , ϕ − Ω I 2 2 γ 2 v 4 / 3 + v 2 / 3 V , ϕ ϕ − κ 2 v 2 / 3 V . (31) m_{\mathrm{eff,raw}}^{2}=-\frac{3\kappa^{2}\gamma^{2}v^{2/3}p_{\phi}^{4}}{8v^{4/3}\Omega_{\mathrm{I}}^{2}}+\frac{5\kappa p_{\phi}^{2}}{2v^{4/3}}+\frac{\kappa^{2}\gamma^{2}v^{2/3}p_{\phi}^{2}}{4\Omega_{\mathrm{I}}^{2}}+\frac{2\kappa\gamma v^{2/3}p_{\phi}\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^{2}}V_{,\phi}-\frac{\Omega_{\mathrm{I}}^{2}}{2\gamma^{2}v^{4/3}}+v^{2/3}V_{,\phi\phi}-\frac{\kappa}{2}v^{2/3}V. \tag*{(31)} m eff , raw 2 = − 8 v 4/3 Ω I 2 3 κ 2 γ 2 v 2/3 p ϕ 4 + 2 v 4/3 5 κ p ϕ 2 + 4 Ω I 2 κ 2 γ 2 v 2/3 p ϕ 2 + Ω I 2 2 κγ v 2/3 p ϕ Λ I V , ϕ − 2 γ 2 v 4/3 Ω I 2 + v 2/3 V , ϕϕ − 2 κ v 2/3 V . (31)
The effective mass used in the hybrid mLQC-I mode analysis [26 , 28 ] is
m e f f 2 = κ p ϕ 2 6 v 4 / 3 ( 19 − 3 κ γ 2 p ϕ 2 Ω I 2 ) + v 2 / 3 [ V , ϕ ϕ + 2 κ γ p ϕ Λ I Ω I 2 V , ϕ − 2 κ 3 V ] . (32) m_{\mathrm{eff}}^{2}=\frac{\kappa p_{\phi}^{2}}{6v^{4/3}}\left(19-\frac{3\kappa\gamma^{2}p_{\phi}^{2}}{\Omega_{\mathrm{I}}^{2}}\right)+v^{2/3}\left[V_{,\phi\phi}+\frac{2\kappa\gamma p_{\phi}\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^{2}}V_{,\phi}-\frac{2\kappa}{3}V\right]. \tag*{(32)} m eff 2 = 6 v 4/3 κ p ϕ 2 ( 19 − Ω I 2 3 κ γ 2 p ϕ 2 ) + v 2/3 [ V , ϕϕ + Ω I 2 2 κγ p ϕ Λ I V , ϕ − 3 2 κ V ] . (32)
The two expressions differ only by the effective background constraint:
m e f f , r a w 2 − m e f f 2 = κ ( 2 Ω I 2 + 3 κ γ 2 p ϕ 2 ) 12 v 1 / 3 Ω I 2 H e f f ( 0 ) . (33) m_{\mathrm{eff,raw}}^{2}-m_{\mathrm{eff}}^{2}=\frac{\kappa\left(2\Omega_{\mathrm{I}}^{2}+3\kappa\gamma^{2}p_{\phi}^{2}\right)}{12v^{1/3}\Omega_{\mathrm{I}}^{2}}H_{\mathrm{eff}}^{(0)}. \tag*{(33)} m eff , raw 2 − m eff 2 = 12 v 1/3 Ω I 2 κ ( 2 Ω I 2 + 3 κ γ 2 p ϕ 2 ) H eff ( 0 ) . (33)
Consequently, once the effective background constraint H e f f ( 0 ) = 0 H_{\mathrm{eff}}^{(0)}=0 H eff ( 0 ) = 0 is imposed after quantization, the off-shell canonical construction reproduces precisely the hybrid effective mass. This provides an important consistency check on the ordering of the procedure. Thus, the quadratic effective Hamiltonian reads
H e f f ( 2 ) = N 2 v 1 / 3 ∫ d 3 x [ π ν 2 + ( ∂ ν ) 2 + m e f f 2 ν 2 ] . (34) H_{\mathrm{eff}}^{(2)}=\frac{N}{2v^{1/3}}\int\mathrm{d}^{3}x\left[\pi_{\nu}^{2}+(\partial\nu)^{2}+m_{\mathrm{eff}}^{2}\nu^{2}\right]. \tag*{(34)} H eff ( 2 ) = 2 v 1/3 N ∫ d 3 x [ π ν 2 + ( ∂ ν ) 2 + m eff 2 ν 2 ] . (34)
For conformal time, N = a = v 1 / 3 N=a=v^{1/3} N = a = v 1/3 , Hamilton’s equations give
ν ′ = π ν , π ν ′ = ∂ 2 ν − m e f f 2 ν . (35) \nu'=\pi_{\nu},\qquad\pi_{\nu}'=\partial^{2}\nu-m_{\mathrm{eff}}^{2}\nu. \tag*{(35)} ν ′ = π ν , π ν ′ = ∂ 2 ν − m eff 2 ν . (35)
Thus, each Fourier mode satisfies
ν k ′ ′ + ( k 2 + m e f f 2 ) ν k = 0. (36) \nu_{k}''+\left(k^{2}+m_{\mathrm{eff}}^{2}\right)\nu_{k}=0. \tag*{(36)} ν k ′′ + ( k 2 + m eff 2 ) ν k = 0. (36)
Here and below, the subscript k k k labels the Fourier mode of the corresponding quantity.
Cubic extension of the hybrid prescription
At cubic order one encounters odd inverse powers of the background momentum that are not explicitly required in the quadratic theory. Rather than introducing an independent polymerization prescription for each such power, we adopt the same factor-ordering and superselection-sector prescription used in the hybrid quantization of the mLQC-I background and its perturbations [26 , 40 ] .
For convenience, we define the background operators
X ^ I ≡ Ω ^ I 2 , L ^ I ≡ Λ ^ I , (37) \widehat{X}_{\mathrm{I}}\equiv\widehat{\Omega}_{\mathrm{I}}^{2},\qquad\widehat{L}_{\mathrm{I}}\equiv\widehat{\Lambda}_{\mathrm{I}}, \tag*{(37)} X I ≡ Ω I 2 , L I ≡ Λ I , (37)
where X ^ I \widehat{X}_{\mathrm{I}} X I is the operator representing the squared background momentum and L ^ I \widehat{L}_{\mathrm{I}} L I is the corresponding odd operator chosen so as to preserve the same step-four superselection sectors; the underlying background-operator construction and its superselection structure are discussed in Refs. [26 , 40 ] .
Guided by these constructions, we extend the same ordering principle to arbitrary odd integer powers. Even integer powers are represented by powers of X ^ I \widehat{X}_{\mathrm{I}} X I , whereas odd integer powers are represented symmetrically as
Ω ^ 2 l + 1 = ∣ X ^ I ∣ l / 2 L ^ I ∣ X ^ I ∣ l / 2 , l ∈ Z . (38) \widehat{\Omega}^{2l+1}=\left|\widehat{X}_{\mathrm{I}}\right|^{l/2}\widehat{L}_{\mathrm{I}}\left|\widehat{X}_{\mathrm{I}}\right|^{l/2},\qquad l\in\mathbb{Z}. \tag*{(38)} Ω 2 l + 1 = X I l /2 L I X I l /2 , l ∈ Z . (38)
Thus, for l = − 1 l=-1 l = − 1 , we take
Ω ^ − 1 = ∣ X ^ I ∣ − 1 / 2 L ^ I ∣ X ^ I ∣ − 1 / 2 . (39) \widehat{\Omega}^{-1}=\left|\widehat{X}_{\mathrm{I}}\right|^{-1/2}\widehat{L}_{\mathrm{I}}\left|\widehat{X}_{\mathrm{I}}\right|^{-1/2}. \tag*{(39)} Ω − 1 = X I − 1/2 L I X I − 1/2 . (39)
For a commuting sharply peaked symbol the odd inverse is Λ I / ∣ X I ∣ \Lambda_{\mathrm{I}}/|X_{\mathrm{I}}| Λ I /∣ X I ∣ , where X I = Ω I 2 X_{\mathrm{I}}=\Omega_{\mathrm{I}}^{2} X I = Ω I 2 . On the positive-density constraint surface, X I = κ γ 2 v 2 ρ / 3 > 0 X_{\mathrm{I}}=\kappa\gamma^{2}v^{2}\rho/3>0 X I = κ γ 2 v 2 ρ /3 > 0 , it reduces to
1 Ω ⟶ Λ I Ω I 2 , (40) \frac{1}{\Omega}\longrightarrow\frac{\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^{2}}, \tag*{(40)} Ω 1 ⟶ Ω I 2 Λ I , (40)
which reproduces the prescription already used in the hybrid mLQC-I construction and in the particle-creation analysis of Ref. [28 ] .
The cubic Hamiltonian additionally contains the inverse odd power Ω − 3 \Omega^{-3} Ω − 3 . Taking l = − 2 l=-2 l = − 2 in the same operator prescription gives
Ω ^ − 3 = ∣ X ^ I ∣ − 1 L ^ I ∣ X ^ I ∣ − 1 . (41) \widehat{\Omega}^{-3}=\left|\widehat{X}_{\mathrm{I}}\right|^{-1}\widehat{L}_{\mathrm{I}}\left|\widehat{X}_{\mathrm{I}}\right|^{-1}. \tag*{(41)} Ω − 3 = X I − 1 L I X I − 1 . (41)
Consequently, for sharply peaked semiclassical background states, we find
1 Ω 3 ⟶ Λ I Ω I 4 , (42) \frac{1}{\Omega^{3}}\longrightarrow\frac{\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^{4}}, \tag*{(42)} Ω 3 1 ⟶ Ω I 4 Λ I , (42)
which fixes the inverse-cubic term appearing in the coefficient A \mathcal{A} A .
The same prescription also fixes the positive odd power of Ω \Omega Ω that occurs in the hybrid momentum combination. Indeed, setting l = 0 l=0 l = 0 we find
Ω ^ = L ^ I , (43) \widehat{\Omega}=\widehat{L}_{\mathrm{I}}, \tag*{(43)} Ω = L I , (43)
and therefore, at the effective level, we obtain
P V I ≡ π V − Λ I γ v 2 / 3 V . (44) \mathcal{P}_{\mathcal{V}_{I}} \equiv\pi_{\mathcal{V}}-\frac{\Lambda_{I}}{\gamma v^{2/3}}\mathcal{V}. \tag*{(44)} P V I ≡ π V − γ v 2/3 Λ I V . (44)
We shall use the shorthand
Γ I ≡ Λ I γ v 2 / 3 , (45) \Gamma_{I}\equiv\frac{\Lambda_{I}}{\gamma v^{2/3}}, \tag*{(45)} Γ I ≡ γ v 2/3 Λ I , (45)
so that P V I = π V − Γ I V \mathcal{P}_{\mathcal{V}_{I}}=\pi_{\mathcal{V}}-\Gamma_{I}\mathcal{V} P V I = π V − Γ I V . Thus, replacements Ω − 1 → Λ I / Ω I 2 \Omega^{-1}\to\Lambda_{I}/\Omega_{I}^{2} Ω − 1 → Λ I / Ω I 2 , Ω − 3 → Λ I / Ω I 4 \Omega^{-3}\to\Lambda_{I}/\Omega_{I}^{4} Ω − 3 → Λ I / Ω I 4 , and Ω → Λ I \Omega\to\Lambda_{I} Ω → Λ I are not treated as independent polymerization choices. They follow from the same odd-power factor-ordering prescription.
A useful organizational rule is to retain the composite shift variable ϑ I \vartheta_{I} ϑ I in factorized form. In particular, products generated by ϑ I 2 \vartheta_{I}^{2} ϑ I 2 should not be expanded and then reinterpreted as independent polymerizations of higher inverse powers. The factors entering ϑ I \vartheta_{I} ϑ I are quantized first according to the preceding prescription, and their products are subsequently formed. This preserves the operator factorization inherited from the classical expression.
We stress that this construction does not establish a unique quantization of the cubic theory. Rather, it is the extension selected by maintaining the same factor ordering and superselection-sector prescription used in the hybrid mLQC-I quantization underlying the quadratic theory [26 , 40 ] .
Effective cubic Hamiltonian
With the preceding prescription, we define
A I = − 3 κ 2 γ 2 p ϕ 3 8 v 2 Λ I Ω I 2 + κ 3 γ 3 p ϕ 5 16 v 2 Λ I Ω I 4 + κ γ p ϕ 4 Λ I Ω I 2 V , ϕ ϕ + 1 6 V , ϕ ϕ ϕ , B I = κ γ p ϕ 4 v 2 / 3 Λ I Ω I 2 , C I = κ 2 γ 2 p ϕ 3 4 v 4 / 3 Ω I 2 , D I = κ γ p ϕ 2 2 v 4 / 3 Λ I Ω I 2 , E I = γ p ϕ 4 v 2 / 3 Λ I Ω I 2 . (46) \begin{aligned}
\mathcal{A}_{I}={}&-\frac{3\kappa^{2}\gamma^{2}p_{\phi}^{3}}{8v^{2}}\frac{\Lambda_{I}}{\Omega_{I}^{2}}+\frac{\kappa^{3}\gamma^{3}p_{\phi}^{5}}{16v^{2}}\frac{\Lambda_{I}}{\Omega_{I}^{4}}+\frac{\kappa\gamma p_{\phi}}{4}\frac{\Lambda_{I}}{\Omega_{I}^{2}}V_{,\phi\phi}+\frac{1}{6}V_{,\phi\phi\phi},\\
\mathcal{B}_{I}={}&\frac{\kappa\gamma p_{\phi}}{4v^{2/3}}\frac{\Lambda_{I}}{\Omega_{I}^{2}},\qquad\mathcal{C}_{I}=\frac{\kappa^{2}\gamma^{2}p_{\phi}^{3}}{4v^{4/3}\Omega_{I}^{2}},\\
\mathcal{D}_{I}={}&\frac{\kappa\gamma p_{\phi}^{2}}{2v^{4/3}}\frac{\Lambda_{I}}{\Omega_{I}^{2}},\qquad\mathcal{E}_{I}=\frac{\gamma p_{\phi}}{4v^{2/3}}\frac{\Lambda_{I}}{\Omega_{I}^{2}}. \tag*{(46)}
\end{aligned} A I = B I = D I = − 8 v 2 3 κ 2 γ 2 p ϕ 3 Ω I 2 Λ I + 16 v 2 κ 3 γ 3 p ϕ 5 Ω I 4 Λ I + 4 κγ p ϕ Ω I 2 Λ I V , ϕϕ + 6 1 V , ϕϕϕ , 4 v 2/3 κγ p ϕ Ω I 2 Λ I , C I = 4 v 4/3 Ω I 2 κ 2 γ 2 p ϕ 3 , 2 v 4/3 κγ p ϕ 2 Ω I 2 Λ I , E I = 4 v 2/3 γ p ϕ Ω I 2 Λ I . (46)
Then, the effective shift variable is
ϑ I , k = 3 κ κ 2 { [ p ϕ 2 v 2 / 3 + γ v 4 / 3 6 Λ I Ω I 2 V , ϕ − κ γ 2 p ϕ 3 12 v 2 / 3 Ω I 2 ] V k + γ p ϕ 6 Λ I Ω I 2 P V I , k } . (47) \vartheta_{I,\mathbf{k}}=\frac{3\kappa}{\kappa^{2}}\left\{\left[\frac{p_{\phi}}{2v^{2/3}}+\frac{\gamma v^{4/3}}{6}\frac{\Lambda_{I}}{\Omega_{I}^{2}}V_{,\phi}-\frac{\kappa\gamma^{2}p_{\phi}^{3}}{12v^{2/3}\Omega_{I}^{2}}\right]\mathcal{V}_{\mathbf{k}}+\frac{\gamma p_{\phi}}{6}\frac{\Lambda_{I}}{\Omega_{I}^{2}}\mathcal{P}_{\mathcal{V}_{I,\mathbf{k}}}\right\}. \tag*{(47)} ϑ I , k = κ 2 3 κ { [ 2 v 2/3 p ϕ + 6 γ v 4/3 Ω I 2 Λ I V , ϕ − 12 v 2/3 Ω I 2 κ γ 2 p ϕ 3 ] V k + 6 γ p ϕ Ω I 2 Λ I P V I , k } . (47)
The resulting effective cubic Hamiltonian takes the compact form
H e f f ( 3 ) = N ∫ d 3 x { A I V 3 + B I V P V I 2 − C I V 2 P V I + B I V ( ∂ V ) 2 − D I [ V 2 ∂ 2 ϑ I + V ∂ i ϑ I ∂ i V ] + 1 v 2 / 3 P V I ∂ i V ∂ i ϑ I + E I V [ ( ∂ i ∂ j ϑ I ) ( ∂ i ∂ j ϑ I ) − ( ∂ 2 ϑ I ) 2 ] } . (48) \begin{aligned}
H_{\mathrm{eff}}^{(3)}=N\int d^{3}x\bigg\{&\mathcal{A}_{I}\mathcal{V}^{3}+\mathcal{B}_{I}\mathcal{V}\mathcal{P}_{\mathcal{V}_{I}}^{2}-\mathcal{C}_{I}\mathcal{V}^{2}\mathcal{P}_{\mathcal{V}_{I}}+\mathcal{B}_{I}\mathcal{V}(\partial\mathcal{V})^{2}-\mathcal{D}_{I}\left[\mathcal{V}^{2}\partial^{2}\vartheta_{I}+\mathcal{V}\partial_{i}\vartheta_{I}\partial^{i}\mathcal{V}\right]\\
&+\frac{1}{v^{2/3}}\mathcal{P}_{\mathcal{V}_{I}}\partial_{i}\mathcal{V}\partial^{i}\vartheta_{I}
+\mathcal{E}_{I}\mathcal{V}\left[(\partial_{i}\partial_{j}\vartheta_{I})(\partial^{i}\partial^{j}\vartheta_{I})-(\partial^{2}\vartheta_{I})^{2}\right]\bigg\}. \tag*{(48)}
\end{aligned} H eff ( 3 ) = N ∫ d 3 x { A I V 3 + B I V P V I 2 − C I V 2 P V I + B I V ( ∂ V ) 2 − D I [ V 2 ∂ 2 ϑ I + V ∂ i ϑ I ∂ i V ] + v 2/3 1 P V I ∂ i V ∂ i ϑ I + E I V [ ( ∂ i ∂ j ϑ I ) ( ∂ i ∂ j ϑ I ) − ( ∂ 2 ϑ I ) 2 ] } . (48)
The full effective Hamiltonian through cubic order is therefore
H e f f = H e f f ( 0 ) + H e f f ( 2 ) + H e f f ( 3 ) + ⋯ . (49) H_{\mathrm{eff}}=H_{\mathrm{eff}}^{(0)}+H_{\mathrm{eff}}^{(2)}+H_{\mathrm{eff}}^{(3)}+\cdots. \tag*{(49)} H eff = H eff ( 0 ) + H eff ( 2 ) + H eff ( 3 ) + ⋯ . (49)
The classical limit is immediate:
Ω I 2 → Ω 2 , Λ I → Ω , H e f f ( 3 ) → H c l ( 3 ) . (50) \Omega_{I}^{2}\to\Omega^{2},\qquad\Lambda_{I}\to\Omega,\qquad H_{\mathrm{eff}}^{(3)}\to H_{\mathrm{cl}}^{(3)}. \tag*{(50)} Ω I 2 → Ω 2 , Λ I → Ω , H eff ( 3 ) → H cl ( 3 ) . (50)
Effective continuation of the nonlinear curvature perturbation
The complete relation in Eq. (2.24) is a classical second-order gauge transformation. By itself it does not define a unique quantum-corrected curvature perturbation in the Planck regime. Nevertheless, for the purpose of extending the same hybrid prescription consistently to all terms that will enter a future pre-bounce and near-bounce non-Gaussianity calculation, we define here a minimal term-by-term effective continuation . The factorization specified below is part of this prescription: in general, effective continuation does not preserve products or inverses. Its derivation from the full second-order effective constraint algebra, identification with a geometrically defined comoving curvature perturbation, and regularity must be established separately. The expressions below apply only where their denominators are nonzero. No new polymerization rule is introduced.
From
z = γ v 1 / 3 p ϕ Ω , (51) z=\frac{\gamma v^{1/3}p_{\phi}}{\Omega}, \tag*{(51)} z = Ω γ v 1/3 p ϕ , (51)
the replacement Ω − 1 → Λ I / Ω I 2 \Omega^{-1}\to\Lambda_{I}/\Omega_{I}^{2} Ω − 1 → Λ I / Ω I 2 gives
z I ≡ γ v 1 / 3 p ϕ Λ I Ω I 2 . (52) z_{I}\equiv\gamma v^{1/3}p_{\phi}\frac{\Lambda_{I}}{\Omega_{I}^{2}}. \tag*{(52)} z I ≡ γ v 1/3 p ϕ Ω I 2 Λ I . (52)
The factors of π a − 1 \pi_a^{-1} π a − 1 and π a − 2 \pi_a^{-2} π a − 2 appearing in the full second-order transformation can be written classically as
1 π a = − κ γ a 6 1 Ω , 1 π a 2 = κ 2 γ 2 a 2 36 1 Ω 2 . (53) \frac{1}{\pi_a}=-\frac{\kappa\gamma a}{6}\frac{1}{\Omega},\qquad\frac{1}{\pi_a^2}=\frac{\kappa^2\gamma^2a^2}{36}\frac{1}{\Omega^2}. \tag*{(53)} π a 1 = − 6 κγ a Ω 1 , π a 2 1 = 36 κ 2 γ 2 a 2 Ω 2 1 . (53)
Their effective continuations are therefore
Π 1 I ≡ ( 1 π a ) e f f = − κ γ a 6 Λ I Ω I 2 , \Pi_{1I}\equiv\left(\frac{1}{\pi_a}\right)_{\mathrm{eff}}=-\frac{\kappa\gamma a}{6}\frac{\Lambda_I}{\Omega_I^2}, Π 1 I ≡ ( π a 1 ) eff = − 6 κγ a Ω I 2 Λ I ,
Π 2 I ≡ ( 1 π a 2 ) e f f = κ 2 γ 2 a 2 36 1 Ω I 2 . (54) \Pi_{2I}\equiv\left(\frac{1}{\pi_a^2}\right)_{\mathrm{eff}}=\frac{\kappa^2\gamma^2a^2}{36}\frac{1}{\Omega_I^2}. \tag*{(54)} Π 2 I ≡ ( π a 2 1 ) eff = 36 κ 2 γ 2 a 2 Ω I 2 1 . (54)
Notice that Π 2 I ≠ ( Π 1 I ) 2 \Pi_{2I}\ne(\Pi_{1I})^2 Π 2 I = ( Π 1 I ) 2 in general. This is intentional: the even inverse power Ω − 2 \Omega^{-2} Ω − 2 is represented by the inverse of X ^ I \widehat{X}_I X I , whereas Ω − 1 \Omega^{-1} Ω − 1 is represented by the sector-preserving odd operator in Eq. (39) . Thus Eq. (54) follows the same even/odd factor-ordering prescription already used in the Hamiltonian.
For the shift contribution we use the already-defined effective variable
χ I ≡ N ϑ I , (55) \chi_I\equiv N\vartheta_I, \tag*{(55)} χ I ≡ N ϑ I , (55)
where ϑ I , k \vartheta_{I,k} ϑ I , k is given in Eq. (47) . Finally, direct continuation of the local coefficient gives
C R , I = − 3 2 γ V , ϕ Λ I 2 p ϕ Ω I 2 + κ γ 2 p ϕ Λ I 2 4 Ω I 4 . (56) C_{R,I}=-\frac{3}{2}\frac{\gamma\mathcal{V}_{,\phi}\Lambda_I}{2p_\phi\Omega_I^2}+\frac{\kappa\gamma^2p_\phi\Lambda_I^2}{4\Omega_I^4}. \tag*{(56)} C R , I = − 2 3 2 p ϕ Ω I 2 γ V , ϕ Λ I + 4 Ω I 4 κ γ 2 p ϕ Λ I 2 . (56)
With these definitions, the minimal effective continuation of the complete second-order curvature perturbation is
R e f f ( x , t ) = − a z I δ ϕ + C R , I ( a z I δ ϕ ) 2 − 3 a 2 κ Π 1 I d d t ( a z I δ ϕ ) 2 − 9 a 4 κ 2 Π 2 I a 2 z I 2 ( δ ϕ ) 2 + 9 a 4 κ 2 Π 2 I a 2 z I 2 ∂ − 2 ∂ i ∂ j ( ∂ i δ ϕ ∂ j δ ϕ ) + 3 a 4 κ Π 1 I a z I ∂ i χ I ∂ i δ ϕ − 3 a 4 κ Π 1 I ∂ − 2 ∂ i ∂ j ( ∂ i χ I ∂ j δ ϕ ) . (57) \mathcal{R}_{\mathrm{eff}}(\mathbf{x},t)=-\frac{a}{z_I}\delta\phi+C_{R,I}\left(\frac{a}{z_I}\delta\phi\right)^2-\frac{3a^2}{\kappa}\Pi_{1I}\frac{d}{dt}\left(\frac{a}{z_I}\delta\phi\right)^2-\frac{9a^4}{\kappa^2}\Pi_{2I}\frac{a^2}{z_I^2}(\delta\phi)^2+\frac{9a^4}{\kappa^2}\Pi_{2I}\frac{a^2}{z_I^2}\partial^{-2}\partial_i\partial_j\left(\partial^i\delta\phi\,\partial^j\delta\phi\right)
+\frac{3a^4}{\kappa}\Pi_{1I}\frac{a}{z_I}\partial_i\chi_I\,\partial^i\delta\phi-\frac{3a^4}{\kappa}\Pi_{1I}\partial^{-2}\partial_i\partial_j\left(\partial^i\chi_I\,\partial^j\delta\phi\right). \tag*{(57)} R eff ( x , t ) = − z I a δ ϕ + C R , I ( z I a δ ϕ ) 2 − κ 3 a 2 Π 1 I d t d ( z I a δ ϕ ) 2 − κ 2 9 a 4 Π 2 I z I 2 a 2 ( δ ϕ ) 2 + κ 2 9 a 4 Π 2 I z I 2 a 2 ∂ − 2 ∂ i ∂ j ( ∂ i δ ϕ ∂ j δ ϕ ) + κ 3 a 4 Π 1 I z I a ∂ i χ I ∂ i δ ϕ − κ 3 a 4 Π 1 I ∂ − 2 ∂ i ∂ j ( ∂ i χ I ∂ j δ ϕ ) . (57)
Equation (3.32) retains precisely the structures that become important away from the late-time super-Hubble regime: the local quadratic piece, the time-derivative term, the two spatial-gradient terms, and the two shift-dependent terms. The spatial derivatives in this equation follow the same physical-FLRW convention as in Eq. (25) ; the scalar χ I = N ϑ I \chi_I=N\vartheta_I χ I = N ϑ I itself is independent of this notational choice. Equation (3.32) is therefore the natural starting point for testing the nonlinear correlation functions in the contracting phase and through the bounce.
For comparison with the late-time calculation performed in the present paper, define
q I ≡ − γ z I . (58) q_I\equiv-\frac{\gamma}{z_I}. \tag*{(58)} q I ≡ − z I γ . (58)
On the expanding classical branch, with all relevant modes super-Hubble and derivative/shift terms negligible relative to the retained local term, Eq. (57) is approximated by
R e f f = q I + C R , I q I 2 + ⋯ . (59) \mathcal{R}_{\mathrm{eff}}=q_I+C_{R,I}q_I^2+\cdots. \tag*{(59)} R eff = q I + C R , I q I 2 + ⋯ . (59)
In the low-curvature limit, Ω I 2 → Ω 2 \Omega_I^2\to\Omega^2 Ω I 2 → Ω 2 and Λ I → Ω \Lambda_I\to\Omega Λ I → Ω , so that z I → z z_I\to z z I → z , Π 1 I → π a − 1 \Pi_{1I}\to\pi_a^{-1} Π 1 I → π a − 1 , Π 2 I → π a − 2 \Pi_{2I}\to\pi_a^{-2} Π 2 I → π a − 2 , χ I → χ \chi_I\to\chi χ I → χ , and Eq. (57) reduces to the complete classical relation (2.24).
We emphasize that Eq. (57) is a minimal effective continuation selected by the same hybrid rules used in the quadratic and cubic Hamiltonians; it is not a derivation of a unique nonlinear quantum gauge transformation. Establishing such uniqueness would require the full second-order effective constraint and gauge-transformation algebra. The present late-time results do not depend on this stronger construction, whereas the full expression above makes explicit which terms must be retained in a future pre-bounce analysis.
Before using H e f f ( 3 ) H_{\mathrm{eff}}^{(3)} H eff ( 3 ) as the interaction Hamiltonian, it is useful to state the assumptions under which the operator prescription is formally symmetric and to distinguish density conventions from dependence on the infrared regulator.
We first consider the background operators. In Sec. III C we introduced
X ^ I ≡ Ω ^ I 2 , L ^ I ≡ Λ ^ I , (60) \widehat{X}_I\equiv\widehat{\Omega}_I^2,\qquad\widehat{L}_I\equiv\widehat{\Lambda}_I, \tag*{(60)} X I ≡ Ω I 2 , L I ≡ Λ I , (60)
and represented odd inverse powers by symmetric sandwich orderings. In particular,
Ω ^ I − 1 = ∣ X ^ I ∣ − 1 / 2 L ^ I ∣ X ^ I ∣ − 1 / 2 , (61) \widehat{\Omega}_I^{-1}=\lvert\widehat{X}_I\rvert^{-1/2}\widehat{L}_I\lvert\widehat{X}_I\rvert^{-1/2}, \tag*{(61)} Ω I − 1 = ∣ X I ∣ − 1/2 L I ∣ X I ∣ − 1/2 , (61)
and
Ω ^ I − 3 = ∣ X ^ I ∣ − 1 L ^ I ∣ X ^ I ∣ − 1 . (62) \widehat{\Omega}_I^{-3}=\lvert\widehat{X}_I\rvert^{-1}\widehat{L}_I\lvert\widehat{X}_I\rvert^{-1}. \tag*{(62)} Ω I − 3 = ∣ X I ∣ − 1 L I ∣ X I ∣ − 1 . (62)
The sandwiches are formally symmetric on the assumed common invariant product domain. Their spectral factors use the self-adjoint background realization assumed in Sec. III C, with the inverse kernel excluded. The displayed effective inverse symbols are used on the positive-density constraint surface, not across the kernel or on arbitrary off-shell points.
Products involving powers of the volume and inverse-background operators are ordered symmetrically. For example, a classical product v s F v^{s}F v s F is represented schematically by
v s F ⟶ v ^ s / 2 F ^ v ^ s / 2 , (63) v^{s}F \longrightarrow\hat{v}^{s/2}\hat{F}\hat{v}^{s/2}, \tag*{(63)} v s F ⟶ v ^ s /2 F ^ v ^ s /2 , (63)
whenever F ^ \hat{F} F ^ does not commute with v ^ \hat{v} v ^ . The same principle is applied to the background coefficients A I , … , E I \mathcal{A}_{I},\ldots,\mathcal{E}_{I} A I , … , E I .
There is an additional ordering issue in the matter sector whenever p ϕ p_{\phi} p ϕ multiplies a nontrivial function of ϕ \phi ϕ . Such products are taken in symmetric form. In particular,
p ϕ V , ϕ ϕ ⟶ 1 2 { p ^ ϕ , V , ϕ ϕ ( ϕ ^ ) } . (64) p_{\phi}V_{,\phi\phi}\longrightarrow\frac{1}{2}\left\{\hat{p}_{\phi},V_{,\phi\phi}(\hat{\phi})\right\}. \tag*{(64)} p ϕ V , ϕϕ ⟶ 2 1 { p ^ ϕ , V , ϕϕ ( ϕ ^ ) } . (64)
The remaining factors of p ϕ p_{\phi} p ϕ commute with purely geometric operators and therefore do not introduce an additional ordering ambiguity.
The odd background momentum entering the perturbative momentum combination is represented by L ^ I \hat{L}_{I} L ^ I . A symmetric realization of the corresponding background coefficient is therefore
Γ ^ I = 1 γ v ^ − 1 / 3 L ^ I v ^ − 1 / 3 , (65) \hat{\Gamma}_{I}=\frac{1}{\gamma}\hat{v}^{-1/3}\hat{L}_{I}\hat{v}^{-1/3}, \tag*{(65)} Γ ^ I = γ 1 v ^ − 1/3 L ^ I v ^ − 1/3 , (65)
whose sharply peaked effective value is
Γ I = Λ I γ v 2 / 3 . (66) \Gamma_{I}=\frac{\Lambda_{I}}{\gamma v^{2/3}}. \tag*{(66)} Γ I = γ v 2/3 Λ I . (66)
Consequently, we find that
P ^ ν I = π ^ ν − Γ ^ I v ^ (67) \hat{P}_{\nu I}=\hat{\pi}_{\nu}-\hat{\Gamma}_{I}\hat{v} \tag*{(67)} P ^ ν I = π ^ ν − Γ ^ I v ^ (67)
is symmetric, since background operators commute with the perturbative canonical pair.
At cubic order an additional ordering choice is required because ν ^ \hat{\nu} ν ^ and P ^ ν I \hat{P}_{\nu I} P ^ ν I do not commute. We adopt Weyl ordering for the perturbative monomials. For three operators A ^ \hat{A} A ^ , B ^ \hat{B} B ^ , and C ^ \hat{C} C ^ , let ( A ^ 1 , A ^ 2 , A ^ 3 ) = ( A ^ , B ^ , C ^ ) (\hat{A}_{1},\hat{A}_{2},\hat{A}_{3})=(\hat{A},\hat{B},\hat{C}) ( A ^ 1 , A ^ 2 , A ^ 3 ) = ( A ^ , B ^ , C ^ ) and define
W [ A ^ B ^ C ^ ] ≡ 1 3 ! ∑ σ ∈ S 3 A ^ σ ( 1 ) A ^ σ ( 2 ) A ^ σ ( 3 ) . (68) \mathcal{W}[\hat{A}\hat{B}\hat{C}]\equiv\frac{1}{3!}\sum_{\sigma\in S_{3}}\hat{A}_{\sigma(1)}\hat{A}_{\sigma(2)}\hat{A}_{\sigma(3)}. \tag*{(68)} W [ A ^ B ^ C ^ ] ≡ 3 ! 1 σ ∈ S 3 ∑ A ^ σ ( 1 ) A ^ σ ( 2 ) A ^ σ ( 3 ) . (68)
Thus, for example,
ν P ν I 2 ⟶ W [ ν ^ P ^ ν I P ^ ν I ] , (69) \nu P_{\nu I}^{2}\longrightarrow\mathcal{W}\left[\hat{\nu}\hat{P}_{\nu I}\hat{P}_{\nu I}\right], \tag*{(69)} ν P ν I 2 ⟶ W [ ν ^ P ^ ν I P ^ ν I ] , (69)
and
ν 2 P ν I ⟶ W [ ν ^ ν ^ P ^ ν I ] . (70) \nu^{2}P_{\nu I}\longrightarrow\mathcal{W}\left[\hat{\nu}\hat{\nu}\hat{P}_{\nu I}\right]. \tag*{(70)} ν 2 P ν I ⟶ W [ ν ^ ν ^ P ^ ν I ] . (70)
The same prescription is applied to the cubic terms containing the shift variable ϑ I \vartheta_{I} ϑ I .
In Fourier space the latter may be written schematically as
ϑ ^ I , k = 3 κ k 2 [ F ^ ν I ν ^ k + 1 2 { F ^ P I , P ^ ν I , k } ] , k ≠ 0 , (71) \hat{\vartheta}_{I,k}=\frac{3\kappa}{k^{2}}\left[\hat{F}_{\nu I}\hat{\nu}_{k}+\frac{1}{2}\left\{\hat{F}_{P_{I}},\hat{P}_{\nu I,k}\right\}\right],\qquad k\ne0, \tag*{(71)} ϑ ^ I , k = k 2 3 κ [ F ^ ν I ν ^ k + 2 1 { F ^ P I , P ^ ν I , k } ] , k = 0 , (71)
where the background coefficients F ^ ν I \hat{F}_{\nu I} F ^ ν I and F ^ P I \hat{F}_{P_{I}} F ^ P I are themselves defined with the symmetric background ordering described above. The anticommutator symmetrizes the product because F ^ P I \hat{F}_{P_{I}} F ^ P I need not commute with Γ ^ I \hat{\Gamma}_{I} Γ ^ I . Products of the remaining background-dependent factors are likewise symmetrized. Since
ν ^ k † = ν ^ − k , P ^ ν I , k † = P ^ ν I , − k , (72) \hat{\nu}_{k}^{\dagger}=\hat{\nu}_{-k},\qquad\hat{P}_{\nu I,k}^{\dagger}=\hat{P}_{\nu I,-k}, \tag*{(72)} ν ^ k † = ν ^ − k , P ^ ν I , k † = P ^ ν I , − k , (72)
we obtain
ϑ ^ I , k † = ϑ ^ I , − k . (73) \hat{\vartheta}_{I,k}^{\dagger}=\hat{\vartheta}_{I,-k}. \tag*{(73)} ϑ ^ I , k † = ϑ ^ I , − k . (73)
The inverse Laplacian introduces no additional Hermiticity problem, because k − 2 k^{-2} k − 2 is real and the homogeneous mode k = 0 k=0 k = 0 is excluded from the perturbative shift sector.
To specify the ordering of the complete interaction, write the real-space integrand of Eq. (48) as a sum of individual real monomials K α M α K_{\alpha}M_{\alpha} K α M α , expanding each bracket into its separate terms. Here K α K_{\alpha} K α includes the sign and all background factors of that term, while M α M_{\alpha} M α contains its three perturbative factors, including spatial derivatives where present. Represent K α K_{\alpha} K α by the symmetric background operator prescribed above and set
H ^ e f f ( 3 ) ≔ N 2 ∑ α ∫ d 3 x { K ^ α , W [ M ^ α ] } . (74) \hat{H}_{\mathrm{eff}}^{(3)}\coloneqq\frac{N}{2}\sum_{\alpha}\int d^{3}x\left\{\hat{K}_{\alpha},\mathcal{W}[\hat{M}_{\alpha}]\right\}. \tag*{(74)} H ^ eff ( 3 ) : = 2 N α ∑ ∫ d 3 x { K ^ α , W [ M ^ α ] } . (74)
In this expression W \mathcal{W} W permutes the three complete perturbative factors of each monomial; every shift factor is first defined by Eq. (71) . For example, the term B I ν P ν I 2 \mathcal{B}_{I}\nu\mathcal{P}_{\nu I}^{2} B I ν P ν I 2 is represented by 1 2 { B ^ I , W [ ν ^ P ^ ν I P ^ ν I ] } \frac{1}{2}\{\hat{\mathcal{B}}_{I},\mathcal{W}[\hat{\nu}\hat{P}_{\nu I}\hat{P}_{\nu I}]\} 2 1 { B ^ I , W [ ν ^ P ^ ν I P ^ ν I ]} . The outer anticommutator is needed because background coefficients need not commute with the background operators contained in the perturbative momentum or shift. With a real classical lapse and a common invariant operator domain, this prescription gives
H ^ e f f ( 3 ) † = H ^ e f f ( 3 ) (75) \hat{H}_{\mathrm{eff}}^{(3)\dagger}=\hat{H}_{\mathrm{eff}}^{(3)} \tag*{(75)} H ^ eff ( 3 ) † = H ^ eff ( 3 ) (75)
formally on the common dense domain of the constituent operators. Equation (3.50) should be understood as a statement of formal Hermiticity, or symmetry, of the interaction Hamiltonian. We do not require here a separate proof of essential self-adjointness, which would involve a more detailed functional-analytic study and is not needed for the effective and tree-level calculations considered below. The effective vertex used here is obtained by replacing the ordered background factors by their sharply peaked c-number values and neglecting background moments and correlations. Different orderings can differ by commutator terms; we do not claim to calculate these quantum corrections or to prove their smallness for an arbitrary background state.
Classical GR and low-curvature limit
As a further consistency check, we verify explicitly that the cubic effective Hamiltonian reduces to its classical GR form in the low-curvature regime. Define
x ≡ λ b , ∣ x ∣ ≪ 1. (76) x \equiv\lambda b,\qquad|x| \ll1. \tag*{(76)} x ≡ λb , ∣ x ∣ ≪ 1. (76)
Expanding the trigonometric functions gives
sin 2 x = x 2 − x 4 3 + O ( x 6 ) , \sin^2 x = x^2-\frac{x^4}{3}+O(x^6), sin 2 x = x 2 − 3 x 4 + O ( x 6 ) ,
sin 2 ( 2 x ) = 4 x 2 − 16 x 4 3 + O ( x 6 ) . (77) \sin^2(2x)=4x^2-\frac{16x^4}{3}+O(x^6). \tag*{(77)} sin 2 ( 2 x ) = 4 x 2 − 3 16 x 4 + O ( x 6 ) . (77)
Using (27) , one then finds
Ω I 2 = v 2 b 2 [ 1 − ( γ 2 + 4 3 ) x 2 + O ( x 4 ) ] . (78) \Omega_{\mathrm{I}}^2=v^2b^2\left[1-\left(\gamma^2+\frac{4}{3}\right)x^2+O(x^4)\right]. \tag*{(78)} Ω I 2 = v 2 b 2 [ 1 − ( γ 2 + 3 4 ) x 2 + O ( x 4 ) ] . (78)
On the GR branch the effective Hubble rate follows from the background Hamiltonian,
H e f f = sin ( 2 x ) 2 γ λ [ ( 1 + γ 2 ) cos ( 2 x ) − γ 2 ] = b γ [ 1 − ( 2 γ 2 + 8 3 ) x 2 + O ( x 4 ) ] . (79) \begin{aligned}
H_{\mathrm{eff}}=\frac{\sin(2x)}{2\gamma\lambda}\left[(1+\gamma^2)\cos(2x)-\gamma^2\right] \\
&=\frac{b}{\gamma}\left[1-\left(2\gamma^2+\frac{8}{3}\right)x^2+O(x^4)\right]. \tag*{(79)}
\end{aligned} H eff = 2 γ λ sin ( 2 x ) [ ( 1 + γ 2 ) cos ( 2 x ) − γ 2 ] = γ b [ 1 − ( 2 γ 2 + 3 8 ) x 2 + O ( x 4 ) ] . (79)
Thus the canonical variable Ω c a n = v b \Omega_{\mathrm{can}}=vb Ω can = v b and the effective-rate variable Ω H = γ v H e f f \Omega_H=\gamma vH_{\mathrm{eff}} Ω H = γ v H eff agree only at leading order:
Ω H = Ω c a n [ 1 − ( 2 γ 2 + 8 3 ) x 2 + O ( x 4 ) ] . \Omega_H=\Omega_{\mathrm{can}}\left[1-\left(2\gamma^2+\frac{8}{3}\right)x^2+O(x^4)\right]. Ω H = Ω can [ 1 − ( 2 γ 2 + 3 8 ) x 2 + O ( x 4 ) ] .
In the following rate-based comparison, the classical functions A , … , E , P ν , F , G \mathcal{A},\ldots,\mathcal{E},\mathcal{P}_{\nu},F,G A , … , E , P ν , F , G and z z z mean their classical functional forms evaluated at Ω H \Omega_H Ω H . They are comparison functions, not a redefinition of the canonical variable in Sec. II. Expressed in terms of Ω H \Omega_H Ω H , the expansions are
Ω I 2 = Ω H 2 [ 1 + ( 3 γ 2 + 4 ) x 2 + O ( x 4 ) ] , \Omega_{\mathrm{I}}^2=\Omega_H^2\left[1+(3\gamma^2+4)x^2+O(x^4)\right], Ω I 2 = Ω H 2 [ 1 + ( 3 γ 2 + 4 ) x 2 + O ( x 4 ) ] ,
Λ I = Ω H [ 1 + ( 2 γ 2 + 2 ) x 2 + O ( x 4 ) ] . (80) \Lambda_{\mathrm{I}}=\Omega_H\left[1+(2\gamma^2+2)x^2+O(x^4)\right]. \tag*{(80)} Λ I = Ω H [ 1 + ( 2 γ 2 + 2 ) x 2 + O ( x 4 ) ] . (80)
It follows that
1 Ω I 2 = 1 Ω H 2 [ 1 − ( 3 γ 2 + 4 ) x 2 + O ( x 4 ) ] , \frac{1}{\Omega_{\mathrm{I}}^2}=\frac{1}{\Omega_H^2}\left[1-(3\gamma^2+4)x^2+O(x^4)\right], Ω I 2 1 = Ω H 2 1 [ 1 − ( 3 γ 2 + 4 ) x 2 + O ( x 4 ) ] ,
Λ I Ω I 2 = 1 Ω H [ 1 − ( γ 2 + 2 ) x 2 + O ( x 4 ) ] , \frac{\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^2}=\frac{1}{\Omega_H}\left[1-(\gamma^2+2)x^2+O(x^4)\right], Ω I 2 Λ I = Ω H 1 [ 1 − ( γ 2 + 2 ) x 2 + O ( x 4 ) ] ,
Λ I Ω I 4 = 1 Ω H 3 [ 1 − ( 4 γ 2 + 6 ) x 2 + O ( x 4 ) ] . (81) \frac{\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^4}=\frac{1}{\Omega_H^3}\left[1-(4\gamma^2+6)x^2+O(x^4)\right]. \tag*{(81)} Ω I 4 Λ I = Ω H 3 1 [ 1 − ( 4 γ 2 + 6 ) x 2 + O ( x 4 ) ] . (81)
Thus, all three background structures appearing in the effective cubic Hamiltonian reduce to their classical counterparts:
1 Ω I 2 ⟶ 1 Ω H 2 , Λ I Ω I 2 ⟶ 1 Ω H , Λ I Ω I 4 ⟶ 1 Ω H 3 . (82) \frac{1}{\Omega_{\mathrm{I}}^2}\longrightarrow\frac{1}{\Omega_H^2},\qquad
\frac{\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^2}\longrightarrow\frac{1}{\Omega_H},\qquad
\frac{\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^4}\longrightarrow\frac{1}{\Omega_H^3}. \tag*{(82)} Ω I 2 1 ⟶ Ω H 2 1 , Ω I 2 Λ I ⟶ Ω H 1 , Ω I 4 Λ I ⟶ Ω H 3 1 . (82)
For later use, we introduce
c 1 ≡ − ( γ 2 + 2 ) , c 2 ≡ − ( 3 γ 2 + 4 ) , c_1\equiv-(\gamma^2+2),\qquad c_2\equiv-(3\gamma^2+4), c 1 ≡ − ( γ 2 + 2 ) , c 2 ≡ − ( 3 γ 2 + 4 ) ,
c 3 ≡ − ( 4 γ 2 + 6 ) . (83) c_3\equiv-(4\gamma^2+6). \tag*{(83)} c 3 ≡ − ( 4 γ 2 + 6 ) . (83)
then the effective cubic coefficients have the expansions
A I = A + x 2 { c 1 [ − 3 κ 2 γ p ϕ 3 8 v 2 Ω H 2 + κ γ p ϕ 4 Ω H V , ϕ ϕ ] + c 3 κ 3 γ 3 p ϕ 5 16 v 2 Ω H 3 } + O ( x 4 ) , \mathcal{A}_{\mathrm{I}}=\mathcal{A}+x^2\left\{c_1\left[-\frac{3\kappa^2\gamma p_\phi^3}{8v^2\Omega_H^2}+\frac{\kappa\gamma p_\phi}{4\Omega_H}V_{,\phi\phi}\right]+c_3\frac{\kappa^3\gamma^3p_\phi^5}{16v^2\Omega_H^3}\right\}+O(x^4), A I = A + x 2 { c 1 [ − 8 v 2 Ω H 2 3 κ 2 γ p ϕ 3 + 4 Ω H κγ p ϕ V , ϕϕ ] + c 3 16 v 2 Ω H 3 κ 3 γ 3 p ϕ 5 } + O ( x 4 ) ,
B I = B [ 1 + c 1 x 2 + O ( x 4 ) ] , \mathcal{B}_{\mathrm{I}}=\mathcal{B}\left[1+c_1x^2+O(x^4)\right], B I = B [ 1 + c 1 x 2 + O ( x 4 ) ] ,
C I = C [ 1 + c 2 x 2 + O ( x 4 ) ] , \mathcal{C}_{\mathrm{I}}=\mathcal{C}\left[1+c_2x^2+O(x^4)\right], C I = C [ 1 + c 2 x 2 + O ( x 4 ) ] ,
D I = D [ 1 + c 1 x 2 + O ( x 4 ) ] , \mathcal{D}_{\mathrm{I}}=\mathcal{D}\left[1+c_1x^2+O(x^4)\right], D I = D [ 1 + c 1 x 2 + O ( x 4 ) ] ,
E I = E [ 1 + c 1 x 2 + O ( x 4 ) ] . (84) \mathcal{E}_{\mathrm{I}}=\mathcal{E}\left[1+c_1x^2+O(x^4)\right]. \tag*{(84)} E I = E [ 1 + c 1 x 2 + O ( x 4 ) ] . (84)
Consequently, we obtain
A I ⟶ A , B I ⟶ B , C I ⟶ C , \mathcal{A}_{\mathrm{I}}\longrightarrow\mathcal{A},\qquad
\mathcal{B}_{\mathrm{I}}\longrightarrow\mathcal{B},\qquad
\mathcal{C}_{\mathrm{I}}\longrightarrow\mathcal{C}, A I ⟶ A , B I ⟶ B , C I ⟶ C ,
D I ⟶ D , E I ⟶ E . (85) \mathcal{D}_{\mathrm{I}}\longrightarrow\mathcal{D},\qquad
\mathcal{E}_{\mathrm{I}}\longrightarrow\mathcal{E}. \tag*{(85)} D I ⟶ D , E I ⟶ E . (85)
The same check must be made for the composite perturbative variables. Using Eqs. (3.19) and (3.55), we find
P ν I = π ν − Ω H γ v 2 / 3 [ 1 + 2 ( γ 2 + 1 ) x 2 + O ( x 4 ) ] ν . (86) \mathcal{P}_{\nu\mathrm{I}}=\pi_\nu-\frac{\Omega_H}{\gamma v^{2/3}}\left[1+2(\gamma^2+1)x^2+O(x^4)\right]\nu. \tag*{(86)} P ν I = π ν − γ v 2/3 Ω H [ 1 + 2 ( γ 2 + 1 ) x 2 + O ( x 4 ) ] ν . (86)
and hence
P ν I = P ν − 2 ( γ 2 + 1 ) x 2 Ω H γ v 2 / 3 ν + O ( x 4 ) . (87) \mathcal{P}_{\nu\mathrm{I}}=\mathcal{P}_\nu-2(\gamma^2+1)x^2\frac{\Omega_H}{\gamma v^{2/3}}\nu+O(x^4). \tag*{(87)} P ν I = P ν − 2 ( γ 2 + 1 ) x 2 γ v 2/3 Ω H ν + O ( x 4 ) . (87)
Therefore, we have
P ν I ⟶ P ν . (88) \mathcal{P}_{\nu\mathrm{I}}\longrightarrow\mathcal{P}_\nu. \tag*{(88)} P ν I ⟶ P ν . (88)
For the shift variable it is useful to write
ϑ I , k = 3 κ k 2 [ F I ν k + G I P ν I , k ] . (89) \vartheta_{\mathrm{I},k}=\frac{3\kappa}{k^2}\left[F_{\mathrm{I}}\nu_k+G_{\mathrm{I}}\mathcal{P}_{\nu\mathrm{I},k}\right]. \tag*{(89)} ϑ I , k = k 2 3 κ [ F I ν k + G I P ν I , k ] . (89)
where
F I = p ϕ 2 v 2 / 3 + γ v 4 / 3 6 Λ I Ω I 2 V , ϕ − κ γ 2 p ϕ 3 12 v 2 / 3 Ω I 2 , F_{\mathrm{I}}=\frac{p_\phi}{2v^{2/3}}+\frac{\gamma v^{4/3}}{6}\frac{\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^2}V_{,\phi}-\frac{\kappa\gamma^2p_\phi^3}{12v^{2/3}\Omega_{\mathrm{I}}^2}, F I = 2 v 2/3 p ϕ + 6 γ v 4/3 Ω I 2 Λ I V , ϕ − 12 v 2/3 Ω I 2 κ γ 2 p ϕ 3 ,
G I = γ p ϕ 6 Λ I Ω I 2 . (90) G_{\mathrm{I}}=\frac{\gamma p_\phi}{6}\frac{\Lambda_{\mathrm{I}}}{\Omega_{\mathrm{I}}^2}. \tag*{(90)} G I = 6 γ p ϕ Ω I 2 Λ I . (90)
Then, (81) implies
F I = F + x 2 [ c 1 γ v 4 / 3 6 Ω H V , ϕ − c 2 κ γ 2 p ϕ 3 12 v 2 / 3 Ω H 2 ] + O ( x 4 ) , F_{\mathrm{I}}=F+x^2\left[c_1\frac{\gamma v^{4/3}}{6\Omega_H}V_{,\phi}-c_2\frac{\kappa\gamma^2p_\phi^3}{12v^{2/3}\Omega_H^2}\right]+O(x^4), F I = F + x 2 [ c 1 6 Ω H γ v 4/3 V , ϕ − c 2 12 v 2/3 Ω H 2 κ γ 2 p ϕ 3 ] + O ( x 4 ) ,
G I = G [ 1 + c 1 x 2 + O ( x 4 ) ] . (91) G_{\mathrm{I}}=G\left[1+c_1x^2+O(x^4)\right]. \tag*{(91)} G I = G [ 1 + c 1 x 2 + O ( x 4 ) ] . (91)
where
F = p ϕ 2 v 2 / 3 + γ v 4 / 3 6 Ω H V , ϕ − κ γ 2 p ϕ 3 12 v 2 / 3 Ω H 2 , G = γ p ϕ 6 Ω H . (92) F=\frac{p_\phi}{2v^{2/3}}+\frac{\gamma v^{4/3}}{6\Omega_H}V_{,\phi}-\frac{\kappa\gamma^2p_\phi^3}{12v^{2/3}\Omega_H^2},\qquad
G=\frac{\gamma p_\phi}{6\Omega_H}. \tag*{(92)} F = 2 v 2/3 p ϕ + 6 Ω H γ v 4/3 V , ϕ − 12 v 2/3 Ω H 2 κ γ 2 p ϕ 3 , G = 6 Ω H γ p ϕ . (92)
Together with Eq. (87) , this gives
ϑ I , k = ϑ k + O ( x 2 ) , ϑ I , k ⟶ ϑ k . (93) \vartheta_{I,k}=\vartheta_k+O(x^2),\qquad\vartheta_{I,k}\longrightarrow\vartheta_k. \tag*{(93)} ϑ I , k = ϑ k + O ( x 2 ) , ϑ I , k ⟶ ϑ k . (93)
Combining Eqs. (85) , (88) , and (93) , every term in Eq. (48) approaches the corresponding term in the classical cubic Hamiltonian Eq. (19) . Hence
H e f f ( 3 ) = H c l ( 3 ) + O ( λ b 2 ) , λ b → 0 , (94) H_{\mathrm{eff}}^{(3)}=H_{\mathrm{cl}}^{(3)}+O(\lambda_b^2),\qquad\lambda_b\to0, \tag*{(94)} H eff ( 3 ) = H cl ( 3 ) + O ( λ b 2 ) , λ b → 0 , (94)
and in particular
H e f f ( 3 ) ⟶ H c l ( 3 ) . (95) H_{\mathrm{eff}}^{(3)}\longrightarrow H_{\mathrm{cl}}^{(3)}. \tag*{(95)} H eff ( 3 ) ⟶ H cl ( 3 ) . (95)
The curvature conversion has the same limit. From Eqs. (52) and (56) , we find
z I ⟶ z , C R , I ⟶ C R , R e f f ⟶ R . (96) z_I\longrightarrow z,\qquad\mathcal{C}_{R,I}\longrightarrow\mathcal{C}_R,\qquad\mathcal{R}_{\mathrm{eff}}\longrightarrow\mathcal{R}. \tag*{(96)} z I ⟶ z , C R , I ⟶ C R , R eff ⟶ R . (96)
Correlation functions and perturbativity diagnostic
Exact finite-momentum cubic vertex
For a momentum triangle
k 1 + k 2 + k 3 = 0 , (97) \mathbf{k}_1+\mathbf{k}_2+\mathbf{k}_3=0, \tag*{(97)} k 1 + k 2 + k 3 = 0 , (97)
we define
d i j ≡ k i ⋅ k j , Q i j ≡ d i j 2 − k i 2 k j 2 . (98) d_{ij}\equiv\mathbf{k}_i\cdot\mathbf{k}_j,\qquad Q_{ij}\equiv d_{ij}^{2}-k_i^{2}k_j^{2}. \tag*{(98)} d ij ≡ k i ⋅ k j , Q ij ≡ d ij 2 − k i 2 k j 2 . (98)
For each leg we introduce the linear interaction-picture quantities
u i ≡ u k i , p i ≡ u i ′ − Γ I u i , y i ≡ F I u i + G I p i . (99) u_i\equiv u_{k_i},\qquad p_i\equiv u_i'-\Gamma_Iu_i,\qquad y_i\equiv F_Iu_i+G_Ip_i. \tag*{(99)} u i ≡ u k i , p i ≡ u i ′ − Γ I u i , y i ≡ F I u i + G I p i . (99)
Then, we have
ϑ i = 3 κ k i 2 y i . (100) \vartheta_i=\frac{3\kappa}{k_i^2}y_i. \tag*{(100)} ϑ i = k i 2 3 κ y i . (100)
After complete symmetrization of the three legs, it is numerically convenient to use the vertex with no divisions by the mode functions:
V 123 = A I u 1 u 2 u 3 + B I 3 ( u 1 p 2 p 3 + u 2 p 3 p 1 + u 3 p 1 p 2 ) − C I 3 ( u 1 u 2 p 3 + u 2 u 3 p 1 + u 3 u 1 p 2 ) + B I 6 ( k 1 2 + k 2 2 + k 3 2 ) u 1 u 2 u 3 + κ D I 2 ( y 1 u 2 u 3 + y 2 u 3 u 1 + y 3 u 1 u 2 ) − κ 2 a 2 ∑ c y c y i k i 2 ( d k i p j u k + d j i p k u j ) + 3 κ 2 E I ∑ c y c Q j k k j 2 k k 2 u i y j y k . (101) \begin{aligned}
\mathcal{V}_{123}={}&\mathcal{A}_Iu_1u_2u_3
+\frac{\mathcal{B}_I}{3}(u_1p_2p_3+u_2p_3p_1+u_3p_1p_2)\\
&-\frac{\mathcal{C}_I}{3}(u_1u_2p_3+u_2u_3p_1+u_3u_1p_2)\\
&+\frac{\mathcal{B}_I}{6}(k_1^2+k_2^2+k_3^2)u_1u_2u_3\\
&+\frac{\kappa\mathcal{D}_I}{2}(y_1u_2u_3+y_2u_3u_1+y_3u_1u_2)\\
&-\frac{\kappa}{2a^2}\sum_{\mathrm{cyc}}\frac{y_i}{k_i^2}
(d_{ki}p_ju_k+d_{ji}p_ku_j)\\
&+3\kappa^2\mathcal{E}_I\sum_{\mathrm{cyc}}
\frac{Q_{jk}}{k_j^2k_k^2}u_iy_jy_k. \tag*{(101)}
\end{aligned} V 123 = A I u 1 u 2 u 3 + 3 B I ( u 1 p 2 p 3 + u 2 p 3 p 1 + u 3 p 1 p 2 ) − 3 C I ( u 1 u 2 p 3 + u 2 u 3 p 1 + u 3 u 1 p 2 ) + 6 B I ( k 1 2 + k 2 2 + k 3 2 ) u 1 u 2 u 3 + 2 κ D I ( y 1 u 2 u 3 + y 2 u 3 u 1 + y 3 u 1 u 2 ) − 2 a 2 κ cyc ∑ k i 2 y i ( d k i p j u k + d j i p k u j ) + 3 κ 2 E I cyc ∑ k j 2 k k 2 Q j k u i y j y k . (101)
where ( i , j , k ) (i,j,k) ( i , j , k ) runs over the cyclic permutations of ( 1 , 2 , 3 ) (1,2,3) ( 1 , 2 , 3 ) . Equation (4.5) is exact at finite momentum; no ultraviolet approximation has been made. Under a constant rephasing u i ↦ e i θ i u i u_i\mapsto e^{i\theta_i}u_i u i ↦ e i θ i u i , both p i p_i p i and y i y_i y i acquire the same phase as u i u_i u i . Every term in V 123 \mathcal{V}_{123} V 123 therefore acquires the common factor e i ( θ 1 + θ 2 + θ 3 ) e^{i(\theta_1+\theta_2+\theta_3)} e i ( θ 1 + θ 2 + θ 3 ) . Its conjugate in Eq. (104) cancels the phase of the three external modes, so the analytic bispectrum is invariant.
In-in bispectrum of the hybrid MS variable
We define
⟨ ν ^ k 1 ν ^ k 2 ν ^ k 3 ⟩ = ( 2 π ) 3 δ ( 3 ) ( k 1 + k 2 + k 3 ) B ν ( k 1 , k 2 , k 3 ) . (102) \langle\hat{\nu}_{\mathbf{k}_1}\hat{\nu}_{\mathbf{k}_2}\hat{\nu}_{\mathbf{k}_3}\rangle=(2\pi)^3\delta^{(3)}(\mathbf{k}_1+\mathbf{k}_2+\mathbf{k}_3)B_\nu(k_1,k_2,k_3). \tag*{(102)} ⟨ ν ^ k 1 ν ^ k 2 ν ^ k 3 ⟩ = ( 2 π ) 3 δ ( 3 ) ( k 1 + k 2 + k 3 ) B ν ( k 1 , k 2 , k 3 ) . (102)
For a Gaussian initial state the free three-point function vanishes. To first order in the cubic interaction, we have
⟨ ν 3 ( η ) ⟩ = − i ∫ η 0 η d η ′ ⟨ [ ν 3 ( η ) , H I ( 3 ) ( η ′ ) ] ⟩ . (103) \langle\nu^3(\eta)\rangle=-i\int_{\eta_0}^{\eta}d\eta'\,\left\langle\left[\nu^3(\eta),H_I^{(3)}(\eta')\right]\right\rangle. \tag*{(103)} ⟨ ν 3 ( η )⟩ = − i ∫ η 0 η d η ′ ⟨ [ ν 3 ( η ) , H I ( 3 ) ( η ′ ) ] ⟩ . (103)
Because the completely symmetrized Hamiltonian in Eq. (48) carries no explicit factor 1 / 3 ! 1/3! 1/3 ! , the three external fields can be contracted with the three fields at the vertex in 3 ! = 6 3!=6 3 ! = 6 equivalent ways. Consequently, we find
B ν ( k 1 , k 2 , k 3 ; η ) = 12 Im [ u 1 ( η ) u 2 ( η ) u 3 ( η ) ∫ η 0 η d η ′ a ( η ′ ) V 123 ∗ ( η ′ ) ] . (104) B_\nu(k_1,k_2,k_3;\eta)=12\operatorname{Im}\left[u_1(\eta)u_2(\eta)u_3(\eta)\int_{\eta_0}^{\eta}d\eta'\,a(\eta')\mathcal{V}_{123}^{*}(\eta')\right]. \tag*{(104)} B ν ( k 1 , k 2 , k 3 ; η ) = 12 Im [ u 1 ( η ) u 2 ( η ) u 3 ( η ) ∫ η 0 η d η ′ a ( η ′ ) V 123 ∗ ( η ′ ) ] . (104)
It is useful to define
I 123 ′ ( η ) = 6 a ( η ) V 123 ∗ ( η ) , I 123 ( η 0 ) = 0 , (105) I'_{123}(\eta)=6a(\eta)\mathcal{V}_{123}^{*}(\eta),\qquad I_{123}(\eta_0)=0, \tag*{(105)} I 123 ′ ( η ) = 6 a ( η ) V 123 ∗ ( η ) , I 123 ( η 0 ) = 0 , (105)
so that
B ν ( k 1 , k 2 , k 3 ; η ) = 2 Im [ u 1 ( η ) u 2 ( η ) u 3 ( η ) I 123 ( η ) ] . (106) B_\nu(k_1,k_2,k_3;\eta)=2\operatorname{Im}\left[u_1(\eta)u_2(\eta)u_3(\eta)I_{123}(\eta)\right]. \tag*{(106)} B ν ( k 1 , k 2 , k 3 ; η ) = 2 Im [ u 1 ( η ) u 2 ( η ) u 3 ( η ) I 123 ( η ) ] . (106)
Equation (4.9) allows the in-in integral to be evolved simultaneously with the three linear mode equations, which is particularly convenient for the numerical calculation.
Late-time conversion to the curvature perturbation
The classical nonlinear relation was given in Sec. II D, and its minimal mLQC-I continuation in Sec. III E. We evaluate the correlation functions only on a late super-Hubble surface in the expanding classical regime, where Eq. (96) applies. Accordingly, with z s = a ϕ ˙ / H = z z_s=a\dot{\phi}/H=z z s = a ϕ ˙ / H = z on that surface, the local relation becomes
R = q + C R q 2 + ⋯ , q = − ν z s . (107) \mathcal{R}=q+\mathcal{C}_Rq^2+\cdots,\qquad q=-\frac{\nu}{z_s}. \tag*{(107)} R = q + C R q 2 + ⋯ , q = − z s ν . (107)
where
C R = − 3 2 − V , ϕ 2 H ϕ ˙ + κ ϕ ˙ 2 4 H 2 . (108) C_{\mathcal{R}}=-\frac{3}{2}-\frac{V_{,\phi}}{2H\dot{\phi}}+\frac{\kappa\dot{\phi}^{2}}{4H^{2}}. \tag*{(108)} C R = − 2 3 − 2 H ϕ ˙ V , ϕ + 4 H 2 κ ϕ ˙ 2 . (108)
Spatial-gradient and time-derivative quadratic terms are omitted in Eq. (107) as a local late-time approximation [31 ] . On the late classical branch, the background Klein–Gordon equation gives V , ϕ = − ϕ ¨ − 3 H ϕ ˙ V_{,\phi}=-\ddot{\phi}-3H\dot{\phi} V , ϕ = − ϕ ¨ − 3 H ϕ ˙ . Hence
C R = 1 2 ϕ ¨ H ϕ ˙ + ϵ H 2 , ϵ H ≡ − H ˙ H 2 = κ ϕ ˙ 2 2 H 2 , (109) C_{\mathcal{R}}=\frac{1}{2}\frac{\ddot{\phi}}{H\dot{\phi}}+\frac{\epsilon_H}{2},\qquad\epsilon_H\equiv-\frac{\dot{H}}{H^2}=\frac{\kappa\dot{\phi}^{2}}{2H^2}, \tag*{(109)} C R = 2 1 H ϕ ˙ ϕ ¨ + 2 ϵ H , ϵ H ≡ − H 2 H ˙ = 2 H 2 κ ϕ ˙ 2 , (109)
where the final equality for ϵ H \epsilon_H ϵ H applies in the classical GR regime. In Fourier space,
R k = − ν k z s + C R z s 2 ∫ p ν p ν k − p + ⋯ . (110) \mathcal{R}_{\mathbf{k}}=-\frac{\nu_{\mathbf{k}}}{z_s}+\frac{C_{\mathcal{R}}}{z_s^2}\int_{\mathbf{p}}\nu_{\mathbf{p}}\nu_{\mathbf{k}-\mathbf{p}}+\cdots. \tag*{(110)} R k = − z s ν k + z s 2 C R ∫ p ν p ν k − p + ⋯ . (110)
Because the linear term in Eq. (110) carries a minus sign, the cross term between the linear and quadratic pieces of R \mathcal{R} R carries a minus sign in the retained two-point correction. The Wick term is quadratic in C R C_{\mathcal{R}} C R and is unaffected by this sign.
Retained nonlinear-conversion correction to the curvature power spectrum
We define
⟨ R ^ k R ^ k ′ ⟩ = ( 2 π ) 3 δ ( 3 ) ( k + k ′ ) 2 π 2 k 3 [ P R ( 0 ) ( k ) + Δ P R ( k ) ] . (111) \left\langle\hat{\mathcal{R}}_{\mathbf{k}}\hat{\mathcal{R}}_{\mathbf{k}'}\right\rangle=(2\pi)^3\delta^{(3)}(\mathbf{k}+\mathbf{k}')\frac{2\pi^2}{k^3}\left[\mathcal{P}_{\mathcal{R}}^{(0)}(k)+\Delta\mathcal{P}_{\mathcal{R}}(k)\right]. \tag*{(111)} ⟨ R ^ k R ^ k ′ ⟩ = ( 2 π ) 3 δ ( 3 ) ( k + k ′ ) k 3 2 π 2 [ P R ( 0 ) ( k ) + Δ P R ( k ) ] . (111)
At leading order, we have
P R ( 0 ) ( k ) = k 3 2 π 2 z s 2 ∣ u k ∣ 2 . (112) \mathcal{P}_{\mathcal{R}}^{(0)}(k)=\frac{k^3}{2\pi^2z_s^2}|u_k|^2. \tag*{(112)} P R ( 0 ) ( k ) = 2 π 2 z s 2 k 3 ∣ u k ∣ 2 . (112)
The correction linear in H I ( 3 ) H_I^{(3)} H I ( 3 ) to the ordinary two-point function ⟨ ν ν ⟩ \langle\nu\nu\rangle ⟨ ν ν ⟩ vanishes for a Gaussian interaction-picture state, because it contains an odd number of free fields. At the order retained here, the nonlinear conversion in Eq. (110) therefore gives two contributions: a term involving the tree-level bispectrum and a Gaussian four-point (Wick) term. One obtains
Δ P R ( k ) = − k 3 C R π 2 z s 3 ∫ d 3 p ( 2 π ) 3 B ν ( k , p , − k − p ) + k 3 C R 2 π 2 z s 4 ∫ d 3 p ( 2 π ) 3 ∣ u p ∣ 2 ∣ u ∣ k − p ∣ ∣ 2 . (113) \Delta\mathcal{P}_{\mathcal{R}}(k)=-\frac{k^3C_{\mathcal{R}}}{\pi^2z_s^3}\int\frac{\mathrm{d}^3p}{(2\pi)^3}B_\nu(\mathbf{k},\mathbf{p},-\mathbf{k}-\mathbf{p})
+\frac{k^3C_{\mathcal{R}}^2}{\pi^2z_s^4}\int\frac{\mathrm{d}^3p}{(2\pi)^3}|u_p|^2|u_{|\mathbf{k}-\mathbf{p}|}|^2. \tag*{(113)} Δ P R ( k ) = − π 2 z s 3 k 3 C R ∫ ( 2 π ) 3 d 3 p B ν ( k , p , − k − p ) + π 2 z s 4 k 3 C R 2 ∫ ( 2 π ) 3 d 3 p ∣ u p ∣ 2 ∣ u ∣ k − p ∣ ∣ 2 . (113)
All conversion factors in Eq. (113) are evaluated at the late-time super-Hubble evaluation surface. For numerical purposes, we define
I B ( k ) ≡ ∫ d 3 p ( 2 π ) 3 B ν ( k , p , − k − p ) , I_B(k)\equiv\int\frac{\mathrm{d}^3p}{(2\pi)^3}B_\nu(\mathbf{k},\mathbf{p},-\mathbf{k}-\mathbf{p}), I B ( k ) ≡ ∫ ( 2 π ) 3 d 3 p B ν ( k , p , − k − p ) ,
I W ( k ) ≡ ∫ d 3 p ( 2 π ) 3 ∣ u p ∣ 2 ∣ u ∣ k − p ∣ ∣ 2 . (114) I_W(k)\equiv\int\frac{\mathrm{d}^3p}{(2\pi)^3}|u_p|^2|u_{|\mathbf{k}-\mathbf{p}|}|^2. \tag*{(114)} I W ( k ) ≡ ∫ ( 2 π ) 3 d 3 p ∣ u p ∣ 2 ∣ u ∣ k − p ∣ ∣ 2 . (114)
With these definitions, we find
Δ P R P R ( 0 ) = − 2 C R z s ∣ u k ∣ 2 I B ( k ) + 2 C R 2 z s 2 ∣ u k ∣ 2 I W ( k ) . (115) \frac{\Delta\mathcal{P}_{\mathcal{R}}}{\mathcal{P}_{\mathcal{R}}^{(0)}}=-\frac{2C_{\mathcal{R}}}{z_s|u_k|^2}I_B(k)+\frac{2C_{\mathcal{R}}^2}{z_s^2|u_k|^2}I_W(k). \tag*{(115)} P R ( 0 ) Δ P R = − z s ∣ u k ∣ 2 2 C R I B ( k ) + z s 2 ∣ u k ∣ 2 2 C R 2 I W ( k ) . (115)
Equation (4.19) directly measures the nonlinear correction to the curvature two-point function retained in the present calculation. A closely related correlation-function criterion was used in standard LQC in Ref. [31 ] . Genuine O ( [ H I ( 3 ) ] 2 ) O([H_I^{(3)}]^2) O ([ H I ( 3 ) ] 2 ) corrections to ⟨ ν ν ⟩ \langle\nu\nu\rangle ⟨ ν ν ⟩ are not included here and are of the same perturbative order as the retained conversion terms and require a separate calculation. Quartic interactions and the cubic curvature conversion may also contribute at that order [39 ] .
More explicitly, write the curvature conversion as R = q + R 2 + R 3 + ⋯ \mathcal{R}=q+\mathcal{R}_2+\mathcal{R}_3+\cdots R = q + R 2 + R 3 + ⋯ , where R n \mathcal{R}_n R n has degree n n n in the linear perturbations. At leading Gaussian order the power spectrum is proportional to one power of the linear covariance. The next connected two-point order, proportional to two powers of that covariance, generally includes
⟨ R 2 R 2 ⟩ 0 , 2 ⟨ q R 2 ⟩ H 3 , 2 ⟨ q R 3 ⟩ 0 , ⟨ q q ⟩ H 3 2 , ⟨ q q ⟩ H 4 . (116) \left\langle\mathcal{R}_2\mathcal{R}_2\right\rangle_0,\quad2\left\langle q\mathcal{R}_2\right\rangle_{H_3},\quad2\left\langle q\mathcal{R}_3\right\rangle_0,\quad
\left\langle qq\right\rangle_{H_3^2},\quad\left\langle qq\right\rangle_{H_4}. \tag*{(116)} ⟨ R 2 R 2 ⟩ 0 , 2 ⟨ q R 2 ⟩ H 3 , 2 ⟨ q R 3 ⟩ 0 , ⟨ q q ⟩ H 3 2 , ⟨ q q ⟩ H 4 . (116)
The subscripts specify the interaction insertions, with 0 denoting free Gaussian contractions. Equation (4.17) retains only the first two structures, using the local late-time approximation R 2 = C R q 2 \mathcal{R}_2=C_{\mathcal{R}}q^2 R 2 = C R q 2 . Interaction-insertion order is therefore distinct from the final order in the covariance expansion. No hierarchy suppressing the other structures in Eq. (116) is established by the retained diagnostic alone.
Correlation-function perturbativity criterion
It is useful to separate the two contributions in Eq. (115) by defining
δ B ( k ) = − 2 C R z s ∣ u k ∣ 2 I B ( k ) , (117) \delta_B(k)=-\frac{2C_{\mathcal{R}}}{z_s|u_k|^2}I_B(k), \tag*{(117)} δ B ( k ) = − z s ∣ u k ∣ 2 2 C R I B ( k ) , (117)
δ W ( k ) = 2 C R 2 z s 2 ∣ u k ∣ 2 I W ( k ) . (118) \delta_W(k)=\frac{2C_{\mathcal{R}}^2}{z_s^2|u_k|^2}I_W(k). \tag*{(118)} δ W ( k ) = z s 2 ∣ u k ∣ 2 2 C R 2 I W ( k ) . (118)
The perturbativity diagnostic is then
Ξ ( k ) = ∣ δ B ( k ) + δ W ( k ) ∣ . (119) \Xi(k)=|\delta_B(k)+\delta_W(k)|. \tag*{(119)} Ξ ( k ) = ∣ δ B ( k ) + δ W ( k ) ∣. (119)
In the numerical analysis we shall monitor not only Ξ \Xi Ξ , but also ∣ δ B ∣ |\delta_B| ∣ δ B ∣ and ∣ δ W ∣ |\delta_W| ∣ δ W ∣ separately. This prevents an accidental cancellation between the bispectrum and Wick terms from being mistaken for uniform perturbative suppression.
For the numerical calculation it is also useful to introduce a conservative quantity that does not rely on cancellations within the oscillatory bispectrum convolution. From the triangle inequality we define
∣ δ B ( k ) ∣ a b s = ∣ C R ∣ 2 π 2 k ∣ z s ∣ ∣ u k ∣ 2 ∫ k I R k U V d p p ∫ q − q + d q q ∣ B ν ( k , p , q ) ∣ , (120) \left|\delta_B(k)\right|_{\mathrm{abs}}=\frac{\left|C_{\mathcal{R}}\right|}{2\pi^2 k\left|z_s\right|\left|u_k\right|^2}\int_{k_{\mathrm{IR}}}^{k_{\mathrm{UV}}}\mathrm{d}p\,p\int_{q_-}^{q_+}\mathrm{d}q\,q\left|B_\nu(k,p,q)\right|, \tag*{(120)} ∣ δ B ( k ) ∣ abs = 2 π 2 k ∣ z s ∣ ∣ u k ∣ 2 ∣ C R ∣ ∫ k IR k UV d p p ∫ q − q + d q q ∣ B ν ( k , p , q ) ∣ , (120)
where the finite triangle domain is specified below. The associated conservative perturbativity diagnostic is
Ξ a b s ( k ) = ∣ δ B ( k ) ∣ a b s + ∣ δ W ( k ) ∣ . (121) \Xi_{\mathrm{abs}}(k)=\left|\delta_B(k)\right|_{\mathrm{abs}}+\left|\delta_W(k)\right|. \tag*{(121)} Ξ abs ( k ) = ∣ δ B ( k ) ∣ abs + ∣ δ W ( k ) ∣ . (121)
By construction, we have
Ξ ( k ) ≤ Ξ a b s ( k ) . (122) \Xi(k)\leq\Xi_{\mathrm{abs}}(k). \tag*{(122)} Ξ ( k ) ≤ Ξ abs ( k ) . (122)
Thus, Ξ a b s ≪ 1 \Xi_{\mathrm{abs}}\ll1 Ξ abs ≪ 1 is a sufficient condition for perturbative smallness of the correction retained in the present correlation-function test, even when the signed bispectrum integral is difficult to resolve because of rapid momentum-space oscillations.
Reduction of the momentum convolutions and domain of validity
For an isotropic background the three-dimensional momentum convolutions can be reduced to two-dimensional integrals over momentum triangles. Choose the external momentum k \mathbf{k} k along the z z z axis and write
p ≡ ∣ p ∣ , μ ≡ k ^ ⋅ p ^ , p\equiv\left|\mathbf{p}\right|,\qquad\mu\equiv\hat{\mathbf{k}}\cdot\hat{\mathbf{p}}, p ≡ ∣ p ∣ , μ ≡ k ^ ⋅ p ^ ,
q ≡ ∣ k + p ∣ = k 2 + p 2 + 2 k p μ . (123) q\equiv\left|\mathbf{k}+\mathbf{p}\right|=\sqrt{k^2+p^2+2kp\mu}. \tag*{(123)} q ≡ ∣ k + p ∣ = k 2 + p 2 + 2 k p μ . (123)
Rotational invariance gives
d 3 p ( 2 π ) 3 = p 2 d p d μ 4 π 2 , (124) \frac{\mathrm{d}^3p}{(2\pi)^3}=\frac{p^2\mathrm{d}p\,\mathrm{d}\mu}{4\pi^2}, \tag*{(124)} ( 2 π ) 3 d 3 p = 4 π 2 p 2 d p d μ , (124)
and, since q d q = k p d μ q\,\mathrm{d}q=kp\,\mathrm{d}\mu q d q = k p d μ , the angular range − 1 ≤ μ ≤ 1 -1\leq\mu\leq1 − 1 ≤ μ ≤ 1 is equivalent to the triangle condition
∣ k − p ∣ ≤ q ≤ k + p . (125) \left|k-p\right|\leq q\leq k+p. \tag*{(125)} ∣ k − p ∣ ≤ q ≤ k + p . (125)
The bispectrum convolution can therefore be written as
I B ( k ) = 1 4 π 2 k ∫ 0 ∞ d p p ∫ ∣ k − p ∣ k + p d q q B ν ( k , p , q ) . (126) I_B(k)=\frac{1}{4\pi^2 k}\int_0^\infty\mathrm{d}p\,p\int_{\left|k-p\right|}^{k+p}\mathrm{d}q\,q\,B_\nu(k,p,q). \tag*{(126)} I B ( k ) = 4 π 2 k 1 ∫ 0 ∞ d p p ∫ ∣ k − p ∣ k + p d q q B ν ( k , p , q ) . (126)
For the Wick contribution one may make the change of variables p → − p \mathbf{p}\to-\mathbf{p} p → − p , after which the same triangle variables can be used. Hence
I W ( k ) = 1 4 π 2 k ∫ 0 ∞ d p p ∫ ∣ k − p ∣ k + p d q q ∣ u p ∣ 2 ∣ u q ∣ 2 . (127) I_W(k)=\frac{1}{4\pi^2 k}\int_0^\infty\mathrm{d}p\,p\int_{\left|k-p\right|}^{k+p}\mathrm{d}q\,q\,\left|u_p\right|^2\left|u_q\right|^2. \tag*{(127)} I W ( k ) = 4 π 2 k 1 ∫ 0 ∞ d p p ∫ ∣ k − p ∣ k + p d q q ∣ u p ∣ 2 ∣ u q ∣ 2 . (127)
Equations (126) and (127) are the formal triangle representations of the convolutions entering the present perturbativity test.
There is an important restriction on how these expressions are used in the present calculation. The nonlinear conversion Eq. (107) was simplified by neglecting spatial-gradient and time-derivative terms on a late-time hypersurface for which the relevant modes are well outside the Hubble radius. It would therefore be inconsistent to combine this super-Hubble approximation with an unrestricted integration over arbitrarily large internal momenta, for which the omitted derivative terms need not be negligible. In the numerical implementation we instead restrict all three legs of the momentum triangle to
k I R ≤ k , p , q ≤ k U V , k U V ≪ a f H f , (128) k_{\mathrm{IR}}\leq k,p,q\leq k_{\mathrm{UV}},\qquad k_{\mathrm{UV}}\ll a_fH_f, \tag*{(128)} k IR ≤ k , p , q ≤ k UV , k UV ≪ a f H f , (128)
where η f \eta_f η f is the late-time evaluation surface. The internal integration limits are then
q − ( k , p ) ≡ max { ∣ k − p ∣ , k I R } , q_-(k,p)\equiv\max\{\left|k-p\right|,k_{\mathrm{IR}}\}, q − ( k , p ) ≡ max { ∣ k − p ∣ , k IR } ,
q + ( k , p ) ≡ min { k + p , k U V } , (129) q_+(k,p)\equiv\min\{k+p,k_{\mathrm{UV}}\}, \tag*{(129)} q + ( k , p ) ≡ min { k + p , k UV } , (129)
with no contribution whenever q + ≤ q − q_+\leq q_- q + ≤ q − . The numerical convolutions are therefore
I B ( k ) = 1 4 π 2 k ∫ k I R k U V d p p ∫ q − q + d q q B ν ( k , p , q ) , I_B(k)=\frac{1}{4\pi^2 k}\int_{k_{\mathrm{IR}}}^{k_{\mathrm{UV}}}\mathrm{d}p\,p\int_{q_-}^{q_+}\mathrm{d}q\,q\,B_\nu(k,p,q), I B ( k ) = 4 π 2 k 1 ∫ k IR k UV d p p ∫ q − q + d q q B ν ( k , p , q ) ,
I W ( k ) = 1 4 π 2 k ∫ k I R k U V d p p ∫ q − q + d q q ∣ u p ∣ 2 ∣ u q ∣ 2 . (130) I_W(k)=\frac{1}{4\pi^2 k}\int_{k_{\mathrm{IR}}}^{k_{\mathrm{UV}}}\mathrm{d}p\,p\int_{q_-}^{q_+}\mathrm{d}q\,q\,\left|u_p\right|^2\left|u_q\right|^2. \tag*{(130)} I W ( k ) = 4 π 2 k 1 ∫ k IR k UV d p p ∫ q − q + d q q ∣ u p ∣ 2 ∣ u q ∣ 2 . (130)
Substitution into Eqs. (117) and (118) gives the forms used directly in the numerical calculation,
δ B ( k ) = − C R 2 π 2 k z s ∣ u k ∣ 2 ∫ k I R k U V d p p ∫ q − q + d q q B ν ( k , p , q ) , (131) \delta_B(k)=-\frac{C_{\mathcal{R}}}{2\pi^2 k z_s\left|u_k\right|^2}\int_{k_{\mathrm{IR}}}^{k_{\mathrm{UV}}}\mathrm{d}p\,p\int_{q_-}^{q_+}\mathrm{d}q\,q\,B_\nu(k,p,q), \tag*{(131)} δ B ( k ) = − 2 π 2 k z s ∣ u k ∣ 2 C R ∫ k IR k UV d p p ∫ q − q + d q q B ν ( k , p , q ) , (131)
δ W ( k ) = C R 2 2 π 2 k z s 2 ∣ u k ∣ 2 ∫ k I R k U V d p p ∫ q − q + d q q ∣ u p ∣ 2 ∣ u q ∣ 2 . (132) \delta_W(k)=\frac{C_{\mathcal{R}}^2}{2\pi^2 k z_s^2\left|u_k\right|^2}\int_{k_{\mathrm{IR}}}^{k_{\mathrm{UV}}}\mathrm{d}p\,p\int_{q_-}^{q_+}\mathrm{d}q\,q\left|u_p\right|^2\left|u_q\right|^2. \tag*{(132)} δ W ( k ) = 2 π 2 k z s 2 ∣ u k ∣ 2 C R 2 ∫ k IR k UV d p p ∫ q − q + d q q ∣ u p ∣ 2 ∣ u q ∣ 2 . (132)
Extending the convolution to arbitrarily large internal momentum would require retaining the complete second-order field redefinition rather than its super-Hubble reduction and, in general, treating the ultraviolet definition of the resulting composite operators consistently. We therefore do not interpret dependence on an uncontrolled large-p p p extension of Eqs. (126) and (127) as part of the correlation-function test performed here.
For orientation, the size of the two terms can be estimated on the late-time slow-roll surface. There, C R = O ( ϵ H , ϕ ¨ / ( H ϕ ˙ ) ) C_{\mathcal{R}}=\mathcal{O}(\epsilon_H,\ddot{\phi}/(H\dot{\phi})) C R = O ( ϵ H , ϕ ¨ / ( H ϕ ˙ )) . For comparable momenta k 1 ∼ k 2 ∼ k 3 ∼ k k_1\sim k_2\sim k_3\sim k k 1 ∼ k 2 ∼ k 3 ∼ k , after factoring out the standard momentum dependence of the bispectrum, its dimensionless amplitude scales schematically as
k 6 B R ∼ f N L ( P R ( 0 ) ) 2 . (133) k^6B_{\mathcal{R}}\sim f_{\mathrm{NL}}\left(\mathcal{P}_{\mathcal{R}}^{(0)}\right)^2. \tag*{(133)} k 6 B R ∼ f NL ( P R ( 0 ) ) 2 . (133)
Consequently, up to momentum-shape factors and the convolution measure, the two contributions scale parametrically as
δ B ∼ O ( C R f N L P R ( 0 ) ) (134) \delta_B \sim\mathcal{O}\left(C_{\mathcal{R}}f_{\mathrm{NL}}\mathcal{P}_{\mathcal{R}}^{(0)}\right) \tag*{(134)} δ B ∼ O ( C R f NL P R ( 0 ) ) (134)
δ W ∼ O ( C R 2 P R ( 0 ) ) (135) \delta_W \sim\mathcal{O}\left(C_{\mathcal{R}}^2\mathcal{P}_{\mathcal{R}}^{(0)}\right) \tag*{(135)} δ W ∼ O ( C R 2 P R ( 0 ) ) (135)
Thus, even a comparatively large bispectrum need not invalidate the correlation-function expansion, because the correction to the two-point function contains an additional power of the small scalar power spectrum. The same suppression mechanism was emphasized for standard LQC in Ref. [31 ] . Equations (134) and (135) are only order-of-magnitude estimates; all numerical results below are reported from the full finite-momentum expression (115) , without replacing the bispectrum by an f N L f_{\mathrm{NL}} f NL ansatz.
Initial states and correlation-function perturbativity
We now evaluate the correlation-function diagnostic for two physically distinct initial-state prescriptions while keeping the background dynamics, cubic Hamiltonian, late-time conversion, and momentum-domain prescription fixed.
Common numerical strategy
For either prescription, the linear mode equation is solved once on the required momentum grid and the histories u k ( η ) u_k(\eta) u k ( η ) and u k ′ ( η ) u'_k(\eta) u k ′ ( η ) are stored. For each momentum triangle we then construct the exact vertex V 123 \mathcal{V}_{123} V 123 from this mode bank and integrate Eq. (105) to the late super-Hubble evaluation surface. The bispectrum follows from Eq. (106) , after which the momentum convolutions in Eq. (130) are performed.
The bispectrum convolution is highly oscillatory in the internal momenta. The reported implementation uses a triangle-aligned composite grid and compares two spacings. In parallel we evaluate Ξ a b s \Xi_{\mathrm{abs}} Ξ abs , which is insensitive to cancellations between neighboring oscillations and therefore provides a conservative stability diagnostic.
The control parameters k I R k_{\mathrm{IR}} k IR and k U V k_{\mathrm{UV}} k UV in Eq. (128) are varied explicitly while maintaining the super-Hubble condition k U V ≪ a f H f k_{\mathrm{UV}} \ll a_f H_f k UV ≪ a f H f required by the late-time field redefinition. Strict cutoff independence is not assumed; when a tested window has not reached a plateau we quote the result as a finite-window estimate.
An independently implemented alternative expands the exact vertex into a finite sum of separable radial contractions and uses locally validated oscillatory time quadrature; see Appendix C. Its explicit candidate state and fixed control window distinguish these checks from a reproduction of every numerical table below. Spatial, temporal, interpolation, and initial-state tests are kept separate.
Bounce-defined initial state
Initial state and numerical setup
The first initial-state prescription is specified directly at the bounce,
η 0 = η B . (136) \eta_0 = \eta_B. \tag*{(136)} η 0 = η B . (136)
For the finite momentum windows considered below, the linear mode functions are initialized in the instantaneous quadratic ground state at the bounce. A smooth fourth-order adiabatic completion would be required if this prescription were extended arbitrarily far into the ultraviolet, but it is not needed for the finite-window results reported here. Explicit instantaneous data are u k B = ( 2 ω k B ) − 1 / 2 u_{kB}=(2\omega_{kB})^{-1/2} u k B = ( 2 ω k B ) − 1/2 and u k B ′ = − i ω k B u k B u'_{kB}=-i\omega_{kB}u_{kB} u k B ′ = − i ω k B u k B , with ω k B 2 = k 2 + m e f f 2 ( η B ) \omega_{kB}^2=k^2+m_{\mathrm{eff}}^2(\eta_B) ω k B 2 = k 2 + m eff 2 ( η B ) .
For definiteness we use the Starobinsky potential
V ( ϕ ) = 3 m ⋆ 2 32 π G ( 1 − e − 16 π G / 3 ϕ ) 2 (137) V(\phi)=\frac{3m_\star^2}{32\pi G}\left(1-e^{-\sqrt{16\pi G/3}\,\phi}\right)^2 \tag*{(137)} V ( ϕ ) = 32 π G 3 m ⋆ 2 ( 1 − e − 16 π G /3 ϕ ) 2 (137)
with
m ⋆ = 3.08 × 10 − 6 m P , γ = 0.2375 , (138) m_\star=3.08\times10^{-6}m_{\mathrm{P}},\qquad\gamma=0.2375, \tag*{(138)} m ⋆ = 3.08 × 1 0 − 6 m P , γ = 0.2375 , (138)
and bounce data
a B = 1 , ϕ B = − 1.09 m P , ϕ ˙ B > 0. (139) a_B=1,\qquad\phi_B=-1.09m_{\mathrm{P}},\qquad\dot{\phi}_B>0. \tag*{(139)} a B = 1 , ϕ B = − 1.09 m P , ϕ ˙ B > 0. (139)
For the enlarged-window scan below we evaluate the curvature perturbation at N = ln a = 18 N=\ln a=18 N = ln a = 18 and integrate over
0.01 ≤ p / m P , q / m P ≤ 2.05. (140) 0.01\leq p/m_{\mathrm{P},q/m_{\mathrm{P}}\leq2.05.} \tag*{(140)} 0.01 ≤ p / m P , q / m P ≤ 2.05. (140)
At this time a H ≃ 1.01 × 10 2 m P aH\simeq1.01\times10^2m_{\mathrm{P}} a H ≃ 1.01 × 1 0 2 m P , so that even the largest internal momentum satisfies k U V / ( a H ) ≃ 2.0 × 10 − 2 k_{\mathrm{UV}}/(aH)\simeq2.0\times10^{-2} k UV / ( a H ) ≃ 2.0 × 1 0 − 2 . All internal momenta in this window are therefore super-Hubble.
The momentum integral is evaluated on a triangle-aligned composite grid. A dedicated resolution test at k = 0.3 m P k=0.3m_{\mathrm{P}} k = 0.3 m P gives
Ξ a b s = 3.48 × 10 − 6 , Δ k = 0.005 m P , = 3.53 × 10 − 6 , Δ k = 0.0025 m P . (141) \begin{aligned}
\Xi_{\mathrm{abs}}&=3.48\times10^{-6},\qquad&&\Delta k=0.005m_{\mathrm{P}},\\
&=3.53\times10^{-6},\qquad&&\Delta k=0.0025m_{\mathrm{P}}. \tag*{(141)}
\end{aligned} Ξ abs = 3.48 × 1 0 − 6 , = 3.53 × 1 0 − 6 , Δ k = 0.005 m P , Δ k = 0.0025 m P . (141)
for the smaller control window 0.05 ≤ p / m P , q / m P ≤ 1.05 0.05\leq p/m_{\mathrm{P}},q/m_{\mathrm{P}}\leq1.05 0.05 ≤ p / m P , q / m P ≤ 1.05 . The change based on the rounded values is about 1.44 % 1.44\% 1.44% . Figure 1 displays this refinement test. The quoted Δ k t \Delta k_t Δ k t was measured at k t = 3 , 8 , 16 , 24 m P k_t=3,8,16,24m_{\mathrm{P}} k t = 3 , 8 , 16 , 24 m P , whereas this control window has k t ≤ 2.4 m P k_t\leq2.4m_{\mathrm{P}} k t ≤ 2.4 m P .
Figure 1. Momentum-resolution test at k = 0.3 m P k = 0.3m_{\mathrm{P}} k = 0.3 m P for the bounce-defined state on the control window specified in the text. The two quoted estimates of Ξ a b s \Xi_{\mathrm{abs}} Ξ abs differ by about 1.44%.
Finite-window absolute-integrand diagnostic
The enlarged-window scan is summarized in Table I. Over the five sampled external momenta in 0.1 ≤ k / m P ≤ 0.9 0.1\leq k/m_{\mathrm{P}}\leq0.9 0.1 ≤ k / m P ≤ 0.9 , the absolute-integrand diagnostic is estimated at the few-10 − 6 10^{-6} 1 0 − 6 level. The Wick term is much smaller, of order 10 − 11 10^{-11} 1 0 − 11 , so the diagnostic is dominated by the integral of the absolute value of the bispectrum. The same enlarged-window scan is shown graphically in Figure 2 , which shows estimates of Ξ a b s \Xi_{\mathrm{abs}} Ξ abs below 6 × 10 − 6 6 \times10^{-6} 6 × 1 0 − 6 at the five sampled momenta while the Wick contribution is negligible on this scale.
Figure 2. Enlarged-window scan of the perturbativity diagnostics for the bounce-defined state at N = 18 N = 18 N = 18 . The conservative Ξ a b s \Xi_{\mathrm{abs}} Ξ abs does not use cancellations in the oscillatory bispectrum convolution and is estimated below 6 × 10 − 6 6 \times10^{-6} 6 × 1 0 − 6 at the five sampled momenta. The magnitude Ξ \Xi Ξ of the signed correction is smaller but is more sensitive to momentum resolution. The legend's “signed Ξ \Xi Ξ ” denotes this magnitude.
Table 1. Enlarged-window perturbativity scan at N = 18 N = 18 N = 18 for the bounce-defined state. The internal range is 0.01 ≤ p / m P , q / m P ≤ 2.05 0.01 \leq p/m_{\mathrm{P}},q/m_{\mathrm{P}} \leq2.05 0.01 ≤ p / m P , q / m P ≤ 2.05 .
The rapid sign oscillations of B ν ( k , p , q ) B_{\nu}(k,p,q) B ν ( k , p , q ) make the signed bispectrum integral more demanding numerically than the absolute-integrand diagnostic. Its precise residual value therefore should not yet be interpreted as a high-accuracy prediction. For the exact integrals on the same finite domain, Eq. (122) gives
Ξ ( k ) ≤ Ξ a b s ( k ) . (142) \Xi(k) \leq\Xi_{\mathrm{abs}}(k). \tag*{(142)} Ξ ( k ) ≤ Ξ abs ( k ) . (142)
The numerical estimates of Ξ a b s \Xi_{\mathrm{abs}} Ξ abs at the five sampled momenta are below 6 × 10 − 6 6 \times10^{-6} 6 × 1 0 − 6 .
Late-time and momentum-window checks
At fixed k = 0.3 m P k = 0.3m_{\mathrm{P}} k = 0.3 m P and for the control window 0.05 ≤ p / m P , q / m P ≤ 1.05 0.05 \leq p/m_{\mathrm{P}},q/m_{\mathrm{P}} \leq1.05 0.05 ≤ p / m P , q / m P ≤ 1.05 , the conservative result changes only mildly between N = 16 N = 16 N = 16 and N = 18 N = 18 N = 18 ,
Ξ a b s ( N = 16 ) = 3.55 × 10 − 6 , Ξ a b s ( N = 17 ) = 3.57 × 10 − 6 , Ξ a b s ( N = 18 ) = 3.60 × 10 − 6 . (143) \begin{aligned}
\Xi_{\mathrm{abs}}(N = 16) &= 3.55 \times10^{-6},\\
\Xi_{\mathrm{abs}}(N = 17) &= 3.57 \times10^{-6},\\
\Xi_{\mathrm{abs}}(N = 18) &= 3.60 \times10^{-6}. \tag*{(143)}
\end{aligned} Ξ abs ( N = 16 ) Ξ abs ( N = 17 ) Ξ abs ( N = 18 ) = 3.55 × 1 0 − 6 , = 3.57 × 1 0 − 6 , = 3.60 × 1 0 − 6 . (143)
These values indicate mild variation in this integrated estimate on the tested surfaces. However, they do not by themselves establish pointwise bispectrum stability or bound the omitted derivative terms.
The dependence on the finite momentum window is more important. With k I R = 0.05 m P k_{\mathrm{IR}} = 0.05m_{\mathrm{P}} k IR = 0.05 m P , increasing k U V k_{\mathrm{UV}} k UV from 0.65 m P 0.65m_{\mathrm{P}} 0.65 m P to 2.05 m P 2.05m_{\mathrm{P}} 2.05 m P changes the estimated diagnostic at k = 0.3 m P k = 0.3m_{\mathrm{P}} k = 0.3 m P from approximately 2.94 × 10 − 6 2.94 \times10^{-6} 2.94 × 1 0 − 6 to 3.93 × 10 − 6 3.93 \times10^{-6} 3.93 × 1 0 − 6 . At fixed k U V = 1.05 m P k_{\mathrm{UV}} = 1.05m_{\mathrm{P}} k UV = 1.05 m P , lowering k I R k_{\mathrm{IR}} k IR from 0.10 m P 0.10m_{\mathrm{P}} 0.10 m P to 0.005 m P 0.005m_{\mathrm{P}} 0.005 m P raises it from approximately 2.74 × 10 − 6 2.74 \times10^{-6} 2.74 × 1 0 − 6 to 5.36 × 10 − 6 5.36 \times10^{-6} 5.36 × 1 0 − 6 . The ultraviolet accumulation is flattening over the tested range, whereas the infrared side has not yet reached a clear cutoff-independent plateau. We therefore regard the tabulated absolute-integrand values only as finite-window estimates.
Remote-contracting de Sitter initial state
State selection and contracting-branch adiabaticity
The second prescription is the de Sitter vacuum in the remote contracting branch. In the asymptotic pre-de Sitter regime the positive-frequency mode can be written as
v k d S ( η ) = π 4 k k η H μ ( 2 ) ( k η ) , μ = 1 4 − ϵ Λ . (144) v_k^{\mathrm{dS}}(\eta) = \sqrt{\frac{\pi}{4k}}\sqrt{k\eta}\,H_{\mu}^{(2)}(k\eta), \qquad\mu= \sqrt{\frac{1}{4} - \epsilon_{\Lambda}}. \tag*{(144)} v k dS ( η ) = 4 k π k η H μ ( 2 ) ( k η ) , μ = 4 1 − ϵ Λ . (144)
with the coefficient of H μ ( 1 ) H_{\mu}^{(1)} H μ ( 1 ) set to zero. For the representative potential and background initial data used below, ϵ Λ = O ( 10 − 5 ) \epsilon_{\Lambda}=O(10^{-5}) ϵ Λ = O ( 1 0 − 5 ) .
We impose these data at a finite time sufficiently far in the contracting phase and assess the sensitivity to moving that surface farther into the past. Appendix C 3 supplies a finite-start signed-convolution comparison for an explicitly specified independent candidate.
The physical justification of the remote prescription differs from the usual far-past construction in standard LQC. In the latter case, moving the initial surface sufficiently far into the low-curvature contracting phase can place the relevant modes in an ultraviolet adiabatic regime, whereas at or near the bounce adiabaticity need not select a unique state [31 ] .
That last argument does not carry over to mLQC-I. Its remote contracting branch approaches de Sitter spacetime,
H ⟶ − H Λ , a ( t ) ∝ e − H Λ t , H \longrightarrow-H_{\Lambda}, \qquad a(t) \propto e^{-H_{\Lambda}t}, H ⟶ − H Λ , a ( t ) ∝ e − H Λ t ,
η ≃ 1 a H Λ ⟶ 0 + , t ⟶ − ∞ . (145) \eta\simeq\frac{1}{aH_{\Lambda}} \longrightarrow0^{+}, \qquad t \longrightarrow-\infty. \tag*{(145)} η ≃ a H Λ 1 ⟶ 0 + , t ⟶ − ∞. (145)
Consequently, for every fixed finite comoving wave number, k / ( a ∣ H ∣ ) ≃ k η → 0 k/(a|H|)\simeq k\eta\to0 k / ( a ∣ H ∣ ) ≃ k η → 0 rather than infinity. The remote de Sitter state used here is therefore not justified by a far-past adiabatic short-wavelength limit. Rather, the asymptotic de Sitter structure supplies the relevant state-selection principle [27 ] .
Numerical convergence and initial-time stability
The finite-start tests at t i = − 10 , − 15 , − 20 , − 25 t P t_i=-10,-15,-20,-25\,t_{\mathrm{P}} t i = − 10 , − 15 , − 20 , − 25 t P are an initial investigation of the remote prescription. They should not be interpreted as a far-past adiabatic limit: every fixed finite comoving mode becomes nonadiabatic as t → − ∞ t\to-\infty t → − ∞ .
We supplement them with the independent calculation in Appendix C. It fixes ϵ Λ = 4.085090207021769 × 10 − 5 \epsilon_{\Lambda}=4.085090207021769\times10^{-5} ϵ Λ = 4.085090207021769 × 1 0 − 5 and evaluates the full signed convolution at k = 0.3 m P k=0.3m_{\mathrm{P}} k = 0.3 m P on the smaller control window 0.05 ≤ p / m P , q / m P ≤ 1.05 0.05\le p/m_{\mathrm{P}},q/m_{\mathrm{P}}\le1.05 0.05 ≤ p / m P , q / m P ≤ 1.05 , with triangle inequalities and N f = 18 N_f=18 N f = 18 . The critical post-bounce time integral is evaluated by validated local phase-aware quadrature. The retained result is Ξ ≃ 3.45 × 10 − 4 \Xi\simeq3.45\times10^{-4} Ξ ≃ 3.45 × 1 0 − 4 . On matched momentum grids, fresh starts at t i / t P = − 20 , − 25 , − 30 t_i/t_{\mathrm{P}}=-20,-25,-30 t i / t P = − 20 , − 25 , − 30 change this estimate by at most approximately 0.012 % 0.012\% 0.012% relative to the − 25 -25 − 25 reference. All sixteen critical time panels and their independent final-sample tests pass the stated numerical gates.
Scale-adapted comparison with the bounce-defined state
The remote-state convolution is substantially more oscillatory than the bounce-state one. To compare the two prescriptions without introducing a fixed momentum interval that has
very different meaning for different external scales, we use the scale-adapted numerical control window
0.01 k ≤ p , q ≤ 3 k . (146) 0.01\,k\le p,q\le3k. \tag*{(146)} 0.01 k ≤ p , q ≤ 3 k . (146)
The endpoints in Eq. (146) remain numerical controls, not physical cutoffs. Table II shows the magnitude of the signed correction for both initial states. The bounce-defined result is extremely small, whereas the remote result decreases rapidly as the external momentum moves toward the bounce-curvature scale at the sampled momenta. Since the integration window itself varies with k k k , this trend describes the specified moving-window diagnostic rather than an unrestricted integral. For reference we define
k B = a B R B 6 ≃ 1.60094 m P . (147) k_B=a_B\sqrt{\frac{R_B}{6}}\simeq1.60094\,m_{\mathrm{P}}. \tag*{(147)} k B = a B 6 R B ≃ 1.60094 m P . (147)
The remote signed result is cancellation dominated. A useful way of quantifying the sensitivity to the rapidly varying external phase is to average the correction δ = δ B + δ W \delta=\delta_B+\delta_W δ = δ B + δ W with the leading power spectrum as weight,
δ ˉ ( k ) = ⟨ P R ( 0 ) δ ⟩ p h a s e ⟨ P R ( 0 ) ⟩ p h a s e , Ξ ≡ ∣ δ ˉ ∣ . (148) \bar{\delta}(k)=\frac{\left\langle\mathcal{P}_{\mathcal{R}}^{(0)}\,\delta\right\rangle_{\mathrm{phase}}}{\left\langle\mathcal{P}_{\mathcal{R}}^{(0)}\right\rangle_{\mathrm{phase}}}, \qquad\Xi\equiv|\bar{\delta}|. \tag*{(148)} δ ˉ ( k ) = ⟨ P R ( 0 ) ⟩ phase ⟨ P R ( 0 ) δ ⟩ phase , Ξ ≡ ∣ δ ˉ ∣. (148)
For the remote state we find
Ξ = 5.05 × 10 − 2 , 6.82 × 10 − 3 , 8.88 × 10 − 4 , \Xi=5.05\times10^{-2},\ 6.82\times10^{-3},\ 8.88\times10^{-4}, Ξ = 5.05 × 1 0 − 2 , 6.82 × 1 0 − 3 , 8.88 × 1 0 − 4 ,
9.72 × 10 − 5 , 1.76 × 10 − 5 , (149) 9.72\times10^{-5},\ 1.76\times10^{-5}, \tag*{(149)} 9.72 × 1 0 − 5 , 1.76 × 1 0 − 5 , (149)
for k / k B = 0.1 , 0.3 , 0.5 , 0.7 , 1 k/k_B=0.1,0.3,0.5,0.7,1 k / k B = 0.1 , 0.3 , 0.5 , 0.7 , 1 , respectively. The absolute-integrand quantity Ξ a b s \Xi_{\mathrm{abs}} Ξ abs is not small for the remote prescription, so the small signed result cannot be promoted to a cancellation-independent bound. The small signed averages do not bound pointwise corrections, and numerical convergence of the cancellation-dominated integral requires further tests. The reported trend refers to the stated scale-adapted window and sampled momenta.
Primordial non-Gaussianity and CMB projection
Primordial power spectra
Before discussing the three-point function, it is useful to display the late-time leading-order curvature power spectrum for the two initial-state prescriptions. Because the spectra contain rapid local oscillations, the figures report a local phase average and phase standard deviation. These averages are used only to display the slowly varying envelope; the correlation-function convolutions themselves use the resolved mode functions.
Curvature bispectrum and non-Gaussianity parameter
For the later analysis of the primordial three-point function it is useful to state explicitly the convention used for f N L f_{\mathrm{NL}} f NL . We define
P q ( k ) ≡ ∣ u k z s ∣ 2 (150) P_q(k) \equiv\left|\frac{u_k}{z_s}\right|^2 \tag*{(150)} P q ( k ) ≡ z s u k 2 (150)
and the symmetric quadratic combination
Σ q ( k 1 , k 2 , k 3 ) ≡ P q ( k 1 ) P q ( k 2 ) + P q ( k 1 ) P q ( k 3 ) + P q ( k 2 ) P q ( k 3 ) . (151) \Sigma_q(k_1,k_2,k_3) \equiv P_q(k_1)P_q(k_2)+P_q(k_1)P_q(k_3)+P_q(k_2)P_q(k_3). \tag*{(151)} Σ q ( k 1 , k 2 , k 3 ) ≡ P q ( k 1 ) P q ( k 2 ) + P q ( k 1 ) P q ( k 3 ) + P q ( k 2 ) P q ( k 3 ) . (151)
Equation (110) then gives, at tree level,
B R ( k 1 , k 2 , k 3 ) = − B ν ( k 1 , k 2 , k 3 ) z s 3 + 2 C R Σ q ( k 1 , k 2 , k 3 ) . (152) B_{\mathcal{R}}(k_1,k_2,k_3)=-\frac{B_\nu(k_1,k_2,k_3)}{z_s^3}+2C_{\mathcal{R}}\Sigma_q(k_1,k_2,k_3). \tag*{(152)} B R ( k 1 , k 2 , k 3 ) = − z s 3 B ν ( k 1 , k 2 , k 3 ) + 2 C R Σ q ( k 1 , k 2 , k 3 ) . (152)
The overall minus sign in the intrinsic term follows from the linear convention R ( 1 ) = − ν / z s \mathcal{R}^{(1)}=-\nu/z_s R ( 1 ) = − ν / z s established in Eq. (23) . We define
f N L ( k 1 , k 2 , k 3 ) = − 5 B R ( k 1 , k 2 , k 3 ) 6 Σ q ( k 1 , k 2 , k 3 ) . (153) f_{\mathrm{NL}}(k_1,k_2,k_3)=-\frac{5B_{\mathcal{R}}(k_1,k_2,k_3)}{6\Sigma_q(k_1,k_2,k_3)}. \tag*{(153)} f NL ( k 1 , k 2 , k 3 ) = − 6 Σ q ( k 1 , k 2 , k 3 ) 5 B R ( k 1 , k 2 , k 3 ) . (153)
For equilateral configurations this reduces to
f N L e q ( k ) = − 5 18 B R ( k , k , k ) P q ( k ) 2 . (154) f_{\mathrm{NL}}^{\mathrm{eq}}(k)=-\frac{5}{18}\frac{B_{\mathcal{R}}(k,k,k)}{P_q(k)^2}. \tag*{(154)} f NL eq ( k ) = − 18 5 P q ( k ) 2 B R ( k , k , k ) . (154)
The nonlinear-conversion piece alone contributes f N L c o n v = − ( 5 / 3 ) C R f_{\mathrm{NL}}^{\mathrm{conv}}=-(5/3)C_{\mathcal{R}} f NL conv = − ( 5/3 ) C R , which is of slow-roll order on the late evaluation surface.
Large primordial non-Gaussianity
The small retained correction to the power spectrum does not imply a small pointwise three-point function. Figure 3 provides the corresponding two-point baseline for both initial-state prescriptions. Using Eqs. (152) –(154) , the equilateral bispectrum is highly oscillatory. To characterize its envelope
Figure 3. Late-time leading-order curvature power spectrum for (a) the bounce-defined state and (b) the remote de Sitter state. Markers show the phase-averaged P R ( 0 ) \mathcal{P}_{\mathcal{R}}^{(0)} P R ( 0 ) at the sampled values of k / k B k/k_B k / k B , error bars show one phase standard deviation, and the smooth curve is a shape-preserving interpolation through those computed points. The dashed line is the large-k k k reference set by the last sampled point. The strong infrared enhancement, particularly for the remote state, is a primordial momentum-space feature and should not by itself be identified with an observable CMB enhancement.
we retain the originally reported local RMS values of f N L e q f_{\mathrm{NL}}^{\mathrm{eq}} f NL eq in Table III . The bounce-defined state gives values of order 10 3 10^3 1 0 3 –10 4 10^4 1 0 4 over the displayed range, whereas the remote state gives much larger pointwise values. These numbers should not be interpreted as CMB constraints because the oscillations are much faster than the relevant projection scales. The corresponding equilateral envelopes are shown in Figure 4 ; the figure makes the large separation between the two state prescriptions, as well as the oscillatory-envelope character of the quoted amplitudes, directly visible.
Table III. Phase-RMS equilateral primordial non-Gaussianity. The entries characterize the primordial pointwise envelope, not an observed CMB amplitude.
Figure 4. Equilateral primordial non-Gaussianity for (a) the bounce-defined state and (b) the remote de Sitter state. The markers report local RMS amplitude and maximum sampled ∣ f N L e q ∣ |f_{\mathrm{NL}}^{\mathrm{eq}}| ∣ f NL eq ∣ . The solid and dashed curves are shape-preserving interpolants through those points. These are primordial momentum-space quantities, not observed CMB amplitudes.
Rapid k t k_t k t phase and momentum-triangle dependence
A dense scan of the raw curvature bispectrum shows that the dominant rapid phase is organized by the total momentum
k t ≡ k 1 + k 2 + k 3 . (155) k_t \equiv k_1+k_2+k_3. \tag*{(155)} k t ≡ k 1 + k 2 + k 3 . (155)
Across equilateral, squeezed, flattened, and generic triangles, the measured period is
Δ k t ≃ 5.7 × 10 − 4 m P , k I e f f ≡ Δ k t 2 π ≃ 9.1 × 10 − 5 m P . (156) \Delta k_t \simeq5.7\times10^{-4}m_P,\qquad k_I^{\mathrm{eff}}\equiv\frac{\Delta k_t}{2\pi}\simeq9.1\times10^{-5}m_P. \tag*{(156)} Δ k t ≃ 5.7 × 1 0 − 4 m P , k I eff ≡ 2 π Δ k t ≃ 9.1 × 1 0 − 5 m P . (156)
A convenient numerical representation is therefore
f N L ( k 1 , k 2 , k 3 ) ≃ F e q ( k t ) S f ( k t ; r 1 , r 2 , r 3 ) × sin [ k t k I e f f + φ ( k t ; r 1 , r 2 , r 3 ) ] . (157) f_{\mathrm{NL}}(k_1,k_2,k_3)\simeq F_{\mathrm{eq}}(k_t)S_f(k_t;r_1,r_2,r_3)\times\sin\left[\frac{k_t}{k_I^{\mathrm{eff}}}+\varphi(k_t;r_1,r_2,r_3)\right]. \tag*{(157)} f NL ( k 1 , k 2 , k 3 ) ≃ F eq ( k t ) S f ( k t ; r 1 , r 2 , r 3 ) × sin [ k I eff k t + φ ( k t ; r 1 , r 2 , r 3 ) ] . (157)
where r i = k i / k t r_i=k_i/k_t r i = k i / k t and r 1 + r 2 + r 3 = 1 r_1+r_2+r_3=1 r 1 + r 2 + r 3 = 1 ; a state label on F e q F_{\mathrm{eq}} F eq , S f S_f S f , and φ \varphi φ is suppressed in this equation. The rapidly varying part is controlled by k t k_t k t , whereas these three functions vary slowly on the scale Δ k t \Delta k_t Δ k t . We verified this structure on a grid of 37 triangle shapes at k t = 3 , 8 , 16 , k_t=3,8,16, k t = 3 , 8 , 16 , and 24 m P 24m_P 24 m P . For the bounce-defined state the median variation of the normalized shape amplitude over this scale range is about 16%, and the median variation of the relative phase is about 0.19 rad. The corresponding remote-state values are approximately 24% and 0.20 rad, with larger excursions confined mainly to the most squeezed configurations.
Table 4. Coefficients in the equilateral-envelope fit (6.10).
General fitting functions for the oscillatory template
(157) is intended to describe the general momentum-triangle dependence over the sampled domain, not only the four representative configurations shown in Figure 5 . The latter are used only to display the oscillations visually. The numerical fit uses 37 triangle shapes at each of the four reference values k t = 3 k_t = 3 k t = 3 , 8, 16, and 24 m P 24m_{\mathrm{P}} 24 m P for each initial-state prescription, giving N n o d e = 148 N_{\mathrm{node}} = 148 N node = 148 nodes for each state. The current shape grid reaches r min = 0.05 r_{\min} = 0.05 r m i n = 0.05 ; more strongly squeezed triangles are outside the presently validated interpolation domain.
Figure 5. Resolved primordial f N L f_{\mathrm{NL}} f NL oscillations in a narrow interval around k t = 3 m P k_t = 3m_{\mathrm{P}} k t = 3 m P for (a) the bounce-defined state and (b) the remote de Sitter state. The four configurations are equilateral ( 1 / 3 , 1 / 3 , 1 / 3 ) (1/3, 1/3, 1/3) ( 1/3 , 1/3 , 1/3 ) , squeezed ( 0.475 , 0.475 , 0.05 ) (0.475, 0.475, 0.05) ( 0.475 , 0.475 , 0.05 ) , flattened ( 0.49 , 0.255 , 0.255 ) (0.49, 0.255, 0.255) ( 0.49 , 0.255 , 0.255 ) , and generic ( 0.40 , 0.35 , 0.25 ) (0.40, 0.35, 0.25) ( 0.40 , 0.35 , 0.25 ) , where the triplets denote ( k 1 , k 2 , k 3 ) / k t (k_1,k_2,k_3)/k_t ( k 1 , k 2 , k 3 ) / k t . The nearly common rapid period provides direct numerical evidence that the leading oscillatory phase is controlled primarily by k t = k 1 + k 2 + k 3 k_t = k_1 + k_2 + k_3 k t = k 1 + k 2 + k 3 , while the shape modulates the amplitude and subleading phase.
For the slowly varying equilateral envelope we introduce
x ≡ ln ( k t 8 m P ) (158) x \equiv\ln\left(\frac{k_t}{8m_{\mathrm{P}}}\right) \tag*{(158)} x ≡ ln ( 8 m P k t ) (158)
and use the compact cubic representation
F e q ( s ) ( k t ) = exp [ a 0 ( s ) + a 1 ( s ) x + a 2 ( s ) x 2 + a 3 ( s ) x 3 ] , (159) F_{\mathrm{eq}}^{(s)}(k_t) = \exp\left[a_0^{(s)} + a_1^{(s)}x + a_2^{(s)}x^2 + a_3^{(s)}x^3\right], \tag*{(159)} F eq ( s ) ( k t ) = exp [ a 0 ( s ) + a 1 ( s ) x + a 2 ( s ) x 2 + a 3 ( s ) x 3 ] , (159)
where s = B , R s = B, R s = B , R denotes the bounce-defined and remote de Sitter states, respectively. The coefficients obtained from the four equilateral reference scales are listed in Table IV. The cubic form interpolates the four quoted equilateral amplitudes exactly on 3 ≤ k t / m P ≤ 24 3 \le k_t/m_{\mathrm{P}} \le24 3 ≤ k t / m P ≤ 24 ; it should not be extrapolated outside this interval. Here F e q F_{\mathrm{eq}} F eq denotes the fitted sinusoidal amplitude entering (157) and should not be confused with the phase-RMS equilateral amplitude quoted in Table III . Additional k t k_t k t nodes are required before interpreting (159) as a sub-percent interpolation between the reference scales.
For the general triangle dependence it is useful to impose permutation symmetry explicitly. We therefore introduce the two symmetric shape coordinates
U ≡ 6 ∑ i = 1 3 ( r i − 1 3 ) 2 , V ≡ 27 ∏ i = 1 3 ( r i − 1 3 ) , (160) U \equiv6\sum_{i=1}^{3}\left(r_i-\frac{1}{3}\right)^2,\qquad V \equiv27\prod_{i=1}^{3}\left(r_i-\frac{1}{3}\right), \tag*{(160)} U ≡ 6 i = 1 ∑ 3 ( r i − 3 1 ) 2 , V ≡ 27 i = 1 ∏ 3 ( r i − 3 1 ) , (160)
which both vanish for the equilateral triangle. To organize the amplitude and phase without artificial 2 π 2\pi 2 π branch jumps, we define the complex slow envelope
G ( s ) ( k t ; U , V ) ≡ S f ( s ) ( k t ; U , V ) exp [ i Δ φ ( s ) ( k t ; U , V ) ] , (161) \mathcal{G}^{(s)}(k_t;U,V) \equiv\mathcal{S}_{f}^{(s)}(k_t;U,V)\exp\left[i\Delta\varphi^{(s)}(k_t;U,V)\right], \tag*{(161)} G ( s ) ( k t ; U , V ) ≡ S f ( s ) ( k t ; U , V ) exp [ i Δ φ ( s ) ( k t ; U , V ) ] , (161)
with G ( s ) ( k t ; 0 , 0 ) = 1 \mathcal{G}^{(s)}(k_t;0,0)=1 G ( s ) ( k t ; 0 , 0 ) = 1 and a continuously unwrapped relative phase. The full slow phase in (157) is decomposed as
φ ( s ) ( k t ; U , V ) = φ e q ( s ) ( k t ) + Δ φ ( s ) ( k t ; U , V ) , Δ φ ( s ) ( k t ; 0 , 0 ) = 0. (162) \varphi^{(s)}(k_t;U,V)=\varphi_{\mathrm{eq}}^{(s)}(k_t)+\Delta\varphi^{(s)}(k_t;U,V),
\qquad
\Delta\varphi^{(s)}(k_t;0,0)=0. \tag*{(162)} φ ( s ) ( k t ; U , V ) = φ eq ( s ) ( k t ) + Δ φ ( s ) ( k t ; U , V ) , Δ φ ( s ) ( k t ; 0 , 0 ) = 0. (162)
The dominant equilateral phase is already represented by k t / k t e f f k_t/k_t^{\mathrm{eff}} k t / k t eff in (157) ; the residual φ e q ( s ) ( k t ) \varphi_{\mathrm{eq}}^{(s)}(k_t) φ eq ( s ) ( k t ) varies slowly but is evaluated numerically in the present analysis rather than assigned a global polynomial fit from only four k t k_t k t reference nodes.
For numerical work we first interpolate two unconstrained smooth functions, A ~ ( s ) \widetilde{A}^{(s)} A ( s ) and Φ ~ ( s ) \widetilde{\Phi}^{(s)} Φ ( s ) , in the three variables ( x , U , V ) (x,U,V) ( x , U , V ) using a permutation-invariant thin-plate radial-basis
representation,
A ~ ( s ) ( y ) = p A ( s ) ( y ) + ∑ j = 1 N n o d e w A j ( s ) Φ ( ∣ y − y j ∣ ) , \widetilde{A}^{(s)}(\mathbf{y}) = p_A^{(s)}(\mathbf{y}) + \sum_{j=1}^{N_{\mathrm{node}}} w_{A_j}^{(s)}\Phi(\lvert\mathbf{y}-\mathbf{y}_j\rvert), A ( s ) ( y ) = p A ( s ) ( y ) + j = 1 ∑ N node w A j ( s ) Φ (∣ y − y j ∣) ,
Φ ~ ( s ) ( y ) = p φ ( s ) ( y ) + ∑ j = 1 N n o d e w φ j ( s ) Φ ( ∣ y − y j ∣ ) , (163) \widetilde{\Phi}^{(s)}(\mathbf{y}) = p_{\varphi}^{(s)}(\mathbf{y}) + \sum_{j=1}^{N_{\mathrm{node}}} w_{\varphi_j}^{(s)}\Phi(\lvert\mathbf{y}-\mathbf{y}_j\rvert), \tag*{(163)} Φ ( s ) ( y ) = p φ ( s ) ( y ) + j = 1 ∑ N node w φ j ( s ) Φ (∣ y − y j ∣) , (163)
where
y = ( x , U , V ) , Φ ( ρ ) = ρ 2 ln ρ , Φ ( 0 ) = 0. (164) \mathbf{y}=(x,U,V),\qquad\Phi(\rho)=\rho^2\ln\rho,\qquad\Phi(0)=0. \tag*{(164)} y = ( x , U , V ) , Φ ( ρ ) = ρ 2 ln ρ , Φ ( 0 ) = 0. (164)
The polynomial completion is taken to be linear in ( x , U , V ) (x,U,V) ( x , U , V ) ,
p X ( s ) ( y ) = c X 0 ( s ) + c X x ( s ) x + c X U ( s ) U + c X V ( s ) V , X = A , φ , (165) p_X^{(s)}(\mathbf{y})=c_{X0}^{(s)}+c_{Xx}^{(s)}x+c_{XU}^{(s)}U+c_{XV}^{(s)}V,\qquad X=A,\varphi, \tag*{(165)} p X ( s ) ( y ) = c X 0 ( s ) + c X x ( s ) x + c X U ( s ) U + c X V ( s ) V , X = A , φ , (165)
and the corresponding coefficients are given explicitly in Appendix D.
The radial weights are determined from the standard thin-plate system
\begin{pmatrix}
K+\lambda_X^{(s)}I & P\\
P^T & 0
\end{pmatrix}
\begin{pmatrix}
\mathbf{w}_X^{(s)}\\
\mathbf{c}_X^{(s)}
\end{pmatrix}
=
\begin{pmatrix}
\mathbf{d}_X^{(s)}\\
0
\end{pmatrix}, \tag*{(166)}\end{pmatrix}
where
K i j = Φ ( ∣ y i − y j ∣ ) , P j = ( 1 , x j , U j , V j ) . (167) K_{ij}=\Phi(\lvert\mathbf{y}_i-\mathbf{y}_j\rvert),\qquad P_j=(1,x_j,U_j,V_j). \tag*{(167)} K ij = Φ (∣ y i − y j ∣) , P j = ( 1 , x j , U j , V j ) . (167)
For the amplitude fit, d A , j = ln S f , j d_{A,j}=\ln S_{f,j} d A , j = ln S f , j , while for the phase fit d φ , j = Δ φ j d_{\varphi,j}=\Delta\varphi_j d φ , j = Δ φ j , with both quantities defined relative to the equilateral configuration at the same k t k_t k t . The smoothing parameters used in the fits are
λ A ( B ) = 10 − 5 , λ φ ( B ) = 10 − 5 , \lambda_A^{(B)}=10^{-5},\qquad\lambda_\varphi^{(B)}=10^{-5}, λ A ( B ) = 1 0 − 5 , λ φ ( B ) = 1 0 − 5 ,
λ A ( R ) = 10 − 5 , λ φ ( R ) = 10 − 4 . (168) \lambda_A^{(R)}=10^{-5},\qquad\lambda_\varphi^{(R)}=10^{-4}. \tag*{(168)} λ A ( R ) = 1 0 − 5 , λ φ ( R ) = 1 0 − 4 . (168)
Appendix D summarizes the interpolation prescription.
We then impose the equilateral normalization exactly at every k t k_t k t by defining
ln S f ( s ) ( x , U , V ) = A ~ ( s ) ( x , U , V ) − A ~ ( s ) ( x , 0 , 0 ) , \ln S_f^{(s)}(x,U,V)=\widetilde{A}^{(s)}(x,U,V)-\widetilde{A}^{(s)}(x,0,0), ln S f ( s ) ( x , U , V ) = A ( s ) ( x , U , V ) − A ( s ) ( x , 0 , 0 ) ,
Δ φ ( s ) ( x , U , V ) = Φ ~ ( s ) ( x , U , V ) − Φ ~ ( s ) ( x , 0 , 0 ) . (169) \Delta\varphi^{(s)}(x,U,V)=\widetilde{\Phi}^{(s)}(x,U,V)-\widetilde{\Phi}^{(s)}(x,0,0). \tag*{(169)} Δ φ ( s ) ( x , U , V ) = Φ ( s ) ( x , U , V ) − Φ ( s ) ( x , 0 , 0 ) . (169)
Thus, we have
S f ( s ) ( x , 0 , 0 ) = 1 , Δ φ ( s ) ( x , 0 , 0 ) = 0 (170) S_f^{(s)}(x,0,0)=1,\qquad\Delta\varphi^{(s)}(x,0,0)=0 \tag*{(170)} S f ( s ) ( x , 0 , 0 ) = 1 , Δ φ ( s ) ( x , 0 , 0 ) = 0 (170)
identically, rather than only at the sampled equilateral nodes. The representation therefore treats the four configurations in Figure 5 as ordinary points in a single permutation-invariant fit over the sampled triangle domain.
The 37-shape grid motivates the representation in Eq. (157) , but a common oscillation period does not by itself establish the accuracy of the single-harmonic waveform. The grid does not yet justify a uniform sub-percent interpolation error over the full sampled domain.
Applying a leave-one-node-out test to the normalized RBF prescription above gives median errors
median ∣ δ ln S f ∣ ≃ 5.1 × 10 − 3 ( B ) , 3.64 × 10 − 2 ( R ) , (171) \operatorname{median}\lvert\delta\ln S_f\rvert\simeq5.1\times10^{-3}\ (B),\quad3.64\times10^{-2}\ (R), \tag*{(171)} median ∣ δ ln S f ∣ ≃ 5.1 × 1 0 − 3 ( B ) , 3.64 × 1 0 − 2 ( R ) , (171)
with corresponding 90th-percentile values 3.70 × 10 − 2 3.70\times10^{-2} 3.70 × 1 0 − 2 and 1.44 × 10 − 1 1.44\times10^{-1} 1.44 × 1 0 − 1 . For the relative phase,
median ∣ δ Δ φ ∣ ≃ 6.5 × 10 − 3 r a d ( B ) , 3.3 × 10 − 3 r a d ( R ) , (172) \operatorname{median}\lvert\delta\Delta\varphi\rvert\simeq6.5\times10^{-3}\,\mathrm{rad}\ (B),\quad3.3\times10^{-3}\,\mathrm{rad}\ (R), \tag*{(172)} median ∣ δ Δ φ ∣ ≃ 6.5 × 1 0 − 3 rad ( B ) , 3.3 × 1 0 − 3 rad ( R ) , (172)
while the corresponding 90th-percentile errors are approximately 0.036 and 0.071 rad. The largest interpolation errors remain concentrated toward the squeezed boundary. We therefore distinguish the accuracy of the physical calculation at the directly computed nodes from the interpolation accuracy between them. A genuinely uniform sub-percent template will require an adaptively refined shape and k t k_t k t grid, with the squeezed region sampled more densely.
CMB-projection diagnostics
The observed reduced CMB bispectrum involves the radiation-transfer functions and spherical-Bessel kernels. For temperature anisotropies, we have
b ℓ 1 ℓ 2 ℓ 3 T T T = ( 2 π ) 3 ∫ x 2 d x ∏ i = 1 3 [ ∫ k i 2 d k i ] B R ( k 1 , k 2 , k 3 ) ∏ i = 1 3 [ Δ ℓ i T ( k i ) j ℓ i ( k x ) ] . (173) b_{\ell_1\ell_2\ell_3}^{TTT}=\left(\frac{2}{\pi}\right)^3\int x^2\,\mathrm{d}x\prod_{i=1}^{3}\left[\int k_i^2\,\mathrm{d}k_i\right]B_{\mathcal{R}}(k_1,k_2,k_3)\prod_{i=1}^{3}\left[\Delta_{\ell_i}^{T}(k_i)j_{\ell_i}(kx)\right]. \tag*{(173)} b ℓ 1 ℓ 2 ℓ 3 T T T = ( π 2 ) 3 ∫ x 2 d x i = 1 ∏ 3 [ ∫ k i 2 d k i ] B R ( k 1 , k 2 , k 3 ) i = 1 ∏ 3 [ Δ ℓ i T ( k i ) j ℓ i ( k x ) ] . (173)
Rapid primordial oscillations can be strongly averaged by the CMB projection, whereas smoother bouncing templates can remain observable after transfer projection [37 , 38 ] . The momentum dependence of the primordial bispectrum must therefore be retained rather than reduced to a single pointwise amplitude.
A full CLASS transfer-function calculation is beyond the scope of the present paper. We instead consider two approximate projection diagnostics. First, in a thin-last-scattering-shell approximation we retain the spherical-Bessel kernels and the full numerical shape interpolation F e q S f e i φ F_{\mathrm{eq}}S_fe^{i\varphi} F eq S f e i φ . Second, we replace the purely geometric kernel by the acoustic model
Δ ℓ T ( k ) ∝ e − ( k / k D ) 2 e − k 2 σ ∗ 2 / 2 [ cos ( k r s ) j ℓ ( k D ∗ ) + β sin ( k r s ) j ℓ ′ ( k D ∗ ) ] , (174) \Delta_{\ell}^{T}(k)\propto e^{-(k/k_D)^2}e^{-k^2\sigma_*^2/2}\left[\cos(k r_s)j_{\ell}(kD_*)+\beta\sin(k r_s)j_{\ell}'(kD_*)\right], \tag*{(174)} Δ ℓ T ( k ) ∝ e − ( k / k D ) 2 e − k 2 σ ∗ 2 /2 [ cos ( k r s ) j ℓ ( k D ∗ ) + β sin ( k r s ) j ℓ ′ ( k D ∗ ) ] , (174)
with D ∗ = 1.4 × 10 4 M p c D_*=1.4\times10^4\,\mathrm{Mpc} D ∗ = 1.4 × 1 0 4 Mpc , r s = 144.4 M p c r_s=144.4\,\mathrm{Mpc} r s = 144.4 Mpc , k D = 0.14 M p c − 1 k_D=0.14\,\mathrm{Mpc}^{-1} k D = 0.14 Mpc − 1 , σ ∗ = 15 M p c \sigma_*=15\,\mathrm{Mpc} σ ∗ = 15 Mpc , and baseline β = 0.5 \beta=0.5 β = 0.5 . This model is intended only to test whether ordinary acoustic and damping scales spoil the rapid-phase cancellation; it is not a substitute for a Boltzmann code.
For the oscillatory template it is useful to normalize the projected residual to a phase-unsuppressed absolute-weight reference,
S ℓ 1 ℓ 2 ℓ 3 ≡ ∣ B ℓ 1 ℓ 2 ℓ 3 o s c ∣ ∫ d Π ∣ W ℓ 1 ℓ 2 ℓ 3 B e n v ∣ . (175) S_{\ell_1\ell_2\ell_3} \equiv\frac{\left|B^{\mathrm{osc}}_{\ell_1\ell_2\ell_3}\right|}{\int\mathrm{d}\Pi\left|W_{\ell_1\ell_2\ell_3} B_{\mathrm{env}}\right|}. \tag*{(175)} S ℓ 1 ℓ 2 ℓ 3 ≡ ∫ d Π ∣ W ℓ 1 ℓ 2 ℓ 3 B env ∣ B ℓ 1 ℓ 2 ℓ 3 osc . (175)
where d Π \mathrm{d}\Pi d Π denotes the numerical momentum measure, W W W contains the projection kernels, and B e n v B_{\mathrm{env}} B env is the adopted reference envelope. This normalization avoids artificially large ratios when the approximate acoustic transfer function changes sign.
For an illustrative bounce-to-present conversion, we take 1 m P 1m_{\mathrm{P}} 1 m P at the bounce to correspond to approximately 5 × 10 − 3 M p c − 1 5 \times10^{-3}\,\mathrm{Mpc}^{-1} 5 × 1 0 − 3 Mpc − 1 today. With this adopted mapping, Eq. (156) gives
k I e f f ≃ 4.57 × 10 − 7 M p c − 1 , Δ k t ≃ 2.87 × 10 − 6 M p c − 1 . (176) k_I^{\mathrm{eff}} \simeq4.57 \times10^{-7}\,\mathrm{Mpc}^{-1}, \qquad\Delta k_t \simeq2.87 \times10^{-6}\,\mathrm{Mpc}^{-1}. \tag*{(176)} k I eff ≃ 4.57 × 1 0 − 7 Mpc − 1 , Δ k t ≃ 2.87 × 1 0 − 6 Mpc − 1 . (176)
For this illustrative mapping, the corresponding angular period is much smaller than one multipole: D ∗ Δ k t ≃ 0.040 D_* \Delta k_t \simeq0.040 D ∗ Δ k t ≃ 0.040 when one momentum leg is varied with the others fixed, and approximately 0.013 0.013 0.013 along the common equilateral direction. This is the ultra-rapid regime in which transfer projection can strongly average the primordial phase. For this illustrative mapping, Table V and Figure 6 retain the reported projection outputs.
Table V. Reported acoustic thin-shell projection diagnostic. The quantity S S S is defined in Eq. (175) and is not an observed f N L f_{\mathrm{NL}} f NL .
Figure 6. Baseline acoustic thin-shell projection diagnostic for several representative multipole configurations. The normalization uses the absolute phase-unsuppressed weight in Eq. (175) ; hence the figure measures the cancellation associated with the rapid primordial phase rather than a ratio to a particular standard CMB template.
These illustrative diagnostics are consistent with substantial phase averaging. The numerical residuals and their displayed precision require independent convergence tests and propagation of primordial-template errors. We deliberately stop short of an observational constraint: a quantitative prediction requires actual radiation-transfer functions, a full radial integration, and the same primordial template evaluated over the complete transfer-supported momentum domain.
Discussion and conclusions
In this paper we have constructed and evaluated a retained nonlinear-conversion diagnostic in hybrid mLQC-I. Starting from the classical quadratic and cubic Hamiltonians in a common hybrid canonical representation, we extended the sector-preserving mLQC-I prescription to the cubic theory and constructed the exact finite-momentum interaction vertex. The regular Mukhanov–Sasaki variable v v v is evolved through the quantum-gravity regime, while the curvature perturbation is evaluated on a late super-Hubble surface for the correlation-function calculation. We have also retained the complete classical second-order transformation between δ ϕ \delta\phi δ ϕ and R \mathcal{R} R and given its minimal term-by-term mLQC-I effective continuation. The latter is not needed for the late-time numerical results reported here, but provides the appropriate starting point for a future calculation of nonlinear correlation functions in the contracting phase and through the bounce.
The first main result concerns the retained nonlinear-conversion correction to the scalar two-point function. For the representative bounce-defined initial state, the cancellation-independent finite-window diagnostic satisfies E a b s < 6 × 10 − 6 \mathcal{E}_{\mathrm{abs}} < 6 \times10^{-6} E abs < 6 × 1 0 − 6 at the five sampled external momenta. Thus, the retained nonlinear correction is very small even before making use of oscillatory cancellations. The remote de Sitter initial state is qualitatively different. Its signed correction is highly oscillatory and cancellation dominated; on the common scale-adapted window it decreases from E ≃ 9.9 × 10 − 2 \mathcal{E} \simeq9.9 \times10^{-2} E ≃ 9.9 × 1 0 − 2 at k / k B = 0.1 k/k_{\mathrm{B}} = 0.1 k / k B = 0.1 to E ≃ 2.3 × 10 − 4 \mathcal{E} \simeq2.3 \times10^{-4} E ≃ 2.3 × 1 0 − 4 at k / k B = 1 k/k_{\mathrm{B}} = 1 k / k B = 1 , and the phase-averaged diagnostic shows the same rapid decrease. Because E a b s \mathcal{E}_{\mathrm{abs}} E abs is not small for the remote prescription, its perturbative control is less robust and cannot be promoted to a cancellation-independent statement.
An independent fixed-window calculation provides an additional numerical check of the remote candidate: at k = 0.3 m P k = 0.3m_{\mathrm{P}} k = 0.3 m P and 0.05 ≤ p / m P , q / m P ≤ 1.05 0.05 \leq p/m_{\mathrm{P}},q/m_{\mathrm{P}} \leq1.05 0.05 ≤ p / m P , q / m P ≤ 1.05 , the retained result is E ≃ 3.45 × 10 − 4 \mathcal{E} \simeq3.45 \times10^{-4} E ≃ 3.45 × 1 0 − 4 . Local phase-aware time integration and independent validation overcome the identified critical-interval time aliasing in this conditional test. Its finite-start variation over t i / t P = − 20 , − 25 , − 30 t_i/t_{\mathrm{P}} = -20,-25,-30 t i / t P = − 20 , − 25 , − 30 is at most about 0.012 % 0.012\% 0.012% on matched grids; see Appendix C.
A second central result is that a small correction to the two-point function can coexist with a very large pointwise primordial bispectrum. In the convention used here, the bounce-defined state has a phase-RMS equilateral f N L f_{\mathrm{NL}} f NL of order 10 3 10^{3} 1 0 3 –10 4 10^{4} 1 0 4 over the displayed range, while the remote prescription can be many orders of magnitude larger. This does not contradict the correlation-function perturbativity test: the correction to the two-point function contains an additional power of the small scalar power spectrum, as summarized parametrically in Eqs. (4.38) and (4.39). The reported finite-window calculations illustrate coexistence of a large pointwise tree bispectrum and a small retained nonlinear-conversion diagnostic.
The momentum dependence of this large non-Gaussianity exhibits a simple and useful structure. Over the sampled triangle domain, the dominant rapid phase is organized by the total momentum k t = k 1 + k 2 + k 3 k_t = k_1 + k_2 + k_3 k t = k 1 + k 2 + k 3 , with Δ k t ≃ 5.7 × 10 − 4 m P \Delta k_t \simeq5.7 \times10^{-4}m_{\mathrm{P}} Δ k t ≃ 5.7 × 1 0 − 4 m P , whereas the triangle-shape envelope and the residual phase vary much more slowly. This motivates the general representation in (157) . We have formulated a permutation-invariant fitting framework in terms of a smooth equilateral envelope and a complex shape function.
For the illustrative bounce-to-present mapping adopted in Sec. VI F, the very short primordial phase scale has an immediate consequence for CMB projection. Thin-shell tests retaining the numerical triangle dependence, followed by an acoustic transfer model including monopole, Doppler, diffusion-damping, and finite-width effects, give residual phase-cancellation factors of order 10 − 7 10^{-7} 1 0 − 7 –10 − 6 10^{-6} 1 0 − 6 for representative configurations. Rapid oscillations can be averaged by projection [37 ] , while smoother templates may survive more efficiently [38 ] .
The projection analysis is deliberately a diagnostic rather than an observational prediction. A definitive CMB constraint requires the exact radiation-transfer functions, full radial integration, and a primordial template validated throughout the complete transfer-supported momentum domain [37 ] .
Likewise, the present correlation-function calculation does not establish the absence of every possible strong-coupling or ultraviolet consistency issue [39 ] . Genuine O ( [ H 3 ( 1 ) ] 2 ) \mathcal{O}\left(\left[H_3^{(1)}\right]^2\right) O ( [ H 3 ( 1 ) ] 2 ) loop corrections, quartic and higher interactions, composite-operator renormalization in an unrestricted ultraviolet extension, and any independent effective-field-theory cutoff are logically distinct questions. In addition, the full nonlinear R e f f \mathcal{R}_{\mathrm{eff}} R eff introduced in Sec. III E opens the way to a direct pre-bounce and near-bounce correlation-function analysis in which the time-derivative, gradient, inverse-Laplacian, and shift contributions can no longer be discarded.
Within the retained ABS-type nonlinear-conversion diagnostic, the bounce-state estimates are small at the five tested momenta in the specified finite window. The remote-state signed results are cancellation sensitive and refer to a scale-adapted window. Further checks of the numerical integrals, omitted interaction terms, and primordial template are required before drawing conclusions about the full interacting theory or the precision of the CMB projection.
Acknowledgments
J.S. and R.P. acknowledge the generous support of Baylor University through the Department of Physics and Astronomy graduate program. A.W. is partially supported by the U.S. National Science Foundation under Grant No. PHY-2308845. We also acknowledge the use of ChatGPT (OpenAI) as a computational aid in checking mathematical derivations, assisting with numerical calculations, and developing codes for some of the figures. The authors take full responsibility for the scientific content of this work.
Useful algebraic relations
The relations used repeatedly in converting from ( a , π a ) (a,\pi_a) ( a , π a ) to ( ν , Ω ) (\nu,\Omega) ( ν , Ω ) are
ν = a 3 , π a = − 6 Ω κ γ a ′ (177) \nu= a^3,\qquad\pi_a = -\frac{6\Omega}{\kappa\gamma a'} \tag*{(177)} ν = a 3 , π a = − κγ a ′ 6Ω (177)
which imply
1 π a = − κ γ a 6 Ω , 1 π a 2 = κ 2 γ 2 a 2 36 Ω 2 , (178) \frac{1}{\pi_a}=-\frac{\kappa\gamma a}{6\Omega},\qquad\frac{1}{\pi_a^2}=\frac{\kappa^2\gamma^2a^2}{36\Omega^2}, \tag*{(178)} π a 1 = − 6Ω κγ a , π a 2 1 = 36 Ω 2 κ 2 γ 2 a 2 , (178)
and
1 π a 3 = − κ 3 γ 3 a 3 216 Ω 3 . (179) \frac{1}{\pi_a^3}=-\frac{\kappa^3\gamma^3a^3}{216\Omega^3}. \tag*{(179)} π a 3 1 = − 216 Ω 3 κ 3 γ 3 a 3 . (179)
For the hybrid momentum combination,
δ φ ˉ = a ( π ν + κ π a 6 a ν ) = a π ν − Ω γ a ν = ν 1 / 3 P ν . (180) \delta\bar{\varphi}=a\left(\pi_\nu+\frac{\kappa\pi_a}{6a}\nu\right)=a\pi_\nu-\frac{\Omega}{\gamma a}\nu=\nu^{1/3}\mathcal{P}_\nu. \tag*{(180)} δ φ ˉ = a ( π ν + 6 a κ π a ν ) = a π ν − γ a Ω ν = ν 1/3 P ν . (180)
Canonical completion on the full phase space
In this appendix we verify explicitly that the background-dependent transformation from ( δ ϕ , δ p ϕ o l d ) (\delta\phi,\delta p_{\phi}^{\mathrm{old}}) ( δ ϕ , δ p ϕ old ) to ( ν , π ν ) (\nu,\pi_\nu) ( ν , π ν ) can be completed canonically on the full phase space without generating any additional cubic contribution to the perturbative Hamiltonian.
For a fiducial cell of coordinate volume V 0 V_0 V 0 , we define
⟨ F ⟩ ≡ 1 V 0 ∫ V 0 d 3 x F . (181) \langle F\rangle\equiv\frac{1}{V_0}\int_{V_0}d^3x\,F. \tag*{(181)} ⟨ F ⟩ ≡ V 0 1 ∫ V 0 d 3 x F . (181)
The perturbative transformation can be written as
δ ϕ = ν a , δ p ϕ o l d = a π ν + a K ν , (182) \delta\phi=\frac{\nu}{a},\qquad\delta p_{\phi}^{\mathrm{old}}=a\pi_\nu+a\mathcal{K}\nu, \tag*{(182)} δ ϕ = a ν , δ p ϕ old = a π ν + a K ν , (182)
where
K ≡ κ π a 6 a + 3 p ϕ 2 a 3 π a . (183) \mathcal{K}\equiv\frac{\kappa\pi_a}{6a}+\frac{3p_\phi^2}{a^3\pi_a}. \tag*{(183)} K ≡ 6 a κ π a + a 3 π a 3 p ϕ 2 . (183)
Notice that
δ p ˉ ϕ = δ p ϕ o l d − 3 p ϕ 2 a π a δ ϕ = a ( π ν + κ π a 6 a ν ) , (184) \delta\bar{p}_{\phi}=\delta p_{\phi}^{\mathrm{old}}-\frac{3p_{\phi}^{2}}{a\pi_{a}}\delta\phi=a\left(\pi_{\nu}+\frac{\kappa\pi_{a}}{6a}\nu\right), \tag*{(184)} δ p ˉ ϕ = δ p ϕ old − a π a 3 p ϕ 2 δ ϕ = a ( π ν + 6 a κ π a ν ) , (184)
as used in the main text.
The symplectic potential before the transformation is
Θ = V 0 π a d a + V 0 p ϕ d ϕ + ∫ V 0 d 3 x δ p ϕ o l d d ( δ ϕ ) . (185) \Theta=V_{0}\pi_{a}\,da+V_{0}p_{\phi}\,d\phi+\int_{V_{0}}d^{3}x\,\delta p_{\phi}^{\mathrm{old}}\,d(\delta\phi). \tag*{(185)} Θ = V 0 π a d a + V 0 p ϕ d ϕ + ∫ V 0 d 3 x δ p ϕ old d ( δ ϕ ) . (185)
Substituting Eq. (B.2), integrating an exact differential, and retaining terms through quadratic order in the perturbations gives
Θ = V 0 π a d a + V 0 p ϕ d ϕ + ∫ V 0 d 3 x π ν d ν + V 0 [ Δ π a d a − Δ a d π a + Δ p ϕ d ϕ − Δ ϕ d p ϕ ] + d F 0 + O ( ϵ 4 ) , (186) \begin{aligned}
\Theta={}&V_{0}\pi_{a}\,da+V_{0}p_{\phi}\,d\phi+\int_{V_{0}}d^{3}x\,\pi_{\nu}\,d\nu\\
&+V_{0}\left[\Delta\pi_{a}\,da-\Delta a\,d\pi_{a}+\Delta p_{\phi}\,d\phi-\Delta\phi\,dp_{\phi}\right]\\
&+d\mathcal{F}_{0}+O(\epsilon^{4}), \tag*{(186)}
\end{aligned} Θ = V 0 π a d a + V 0 p ϕ d ϕ + ∫ V 0 d 3 x π ν d ν + V 0 [ Δ π a d a − Δ a d π a + Δ p ϕ d ϕ − Δ ϕ d p ϕ ] + d F 0 + O ( ϵ 4 ) , (186)
where ϵ \epsilon ϵ denotes the perturbative amplitude. The exact differential before completing the homogeneous variables is
F 0 = V 0 2 K ⟨ ν 2 ⟩ . (187) \mathcal{F}_{0}=\frac{V_{0}}{2}\mathcal{K}\langle\nu^{2}\rangle. \tag*{(187)} F 0 = 2 V 0 K ⟨ ν 2 ⟩ . (187)
The required quadratic redefinitions of the homogeneous variables are
Δ a = ( κ 12 a − 3 p ϕ 2 2 a 3 π a 2 ) ⟨ ν 2 ⟩ , (188) \Delta a=\left(\frac{\kappa}{12a}-\frac{3p_{\phi}^{2}}{2a^{3}\pi_{a}^{2}}\right)\langle\nu^{2}\rangle, \tag*{(188)} Δ a = ( 12 a κ − 2 a 3 π a 2 3 p ϕ 2 ) ⟨ ν 2 ⟩ , (188)
Δ ϕ = 3 p ϕ a 3 π a ⟨ ν 2 ⟩ , (189) \Delta\phi=\frac{3p_{\phi}}{a^{3}\pi_{a}}\langle\nu^{2}\rangle, \tag*{(189)} Δ ϕ = a 3 π a 3 p ϕ ⟨ ν 2 ⟩ , (189)
Δ p ϕ = 0 , (190) \Delta p_{\phi}=0, \tag*{(190)} Δ p ϕ = 0 , (190)
and
Δ π a = − 1 a ⟨ ν π ν ⟩ − κ π a 12 a 2 ⟨ ν 2 ⟩ + 3 p ϕ 2 2 a 4 π a ⟨ ν 2 ⟩ . (191) \Delta\pi_{a}=-\frac{1}{a}\langle\nu\pi_{\nu}\rangle-\frac{\kappa\pi_{a}}{12a^{2}}\langle\nu^{2}\rangle+\frac{3p_{\phi}^{2}}{2a^{4}\pi_{a}}\langle\nu^{2}\rangle. \tag*{(191)} Δ π a = − a 1 ⟨ ν π ν ⟩ − 12 a 2 κ π a ⟨ ν 2 ⟩ + 2 a 4 π a 3 p ϕ 2 ⟨ ν 2 ⟩ . (191)
Introducing the completed homogeneous variables
a ~ = a + Δ a , π ~ a = π a + Δ π a , ϕ ~ = ϕ + Δ ϕ , \tilde{a}=a+\Delta a,\qquad\tilde{\pi}_{a}=\pi_{a}+\Delta\pi_{a},\qquad\tilde{\phi}=\phi+\Delta\phi, a ~ = a + Δ a , π ~ a = π a + Δ π a , ϕ ~ = ϕ + Δ ϕ ,
p ~ ϕ = p ϕ + Δ p ϕ (192) \tilde{p}_{\phi}=p_{\phi}+\Delta p_{\phi} \tag*{(192)} p ~ ϕ = p ϕ + Δ p ϕ (192)
we obtain
Θ = V 0 π ~ a d a ~ + V 0 p ~ ϕ d ϕ ~ + ∫ V 0 d 3 x π ν d ν + d F c a n + O ( ϵ 4 ) . (193) \Theta=V_{0}\tilde{\pi}_{a}\,d\tilde{a}+V_{0}\tilde{p}_{\phi}\,d\tilde{\phi}+\int_{V_{0}}d^{3}x\,\pi_{\nu}\,d\nu+d\mathcal{F}_{\mathrm{can}}+O(\epsilon^{4}). \tag*{(193)} Θ = V 0 π ~ a d a ~ + V 0 p ~ ϕ d ϕ ~ + ∫ V 0 d 3 x π ν d ν + d F can + O ( ϵ 4 ) . (193)
The remaining exact differential is determined by
F c a n = F 0 − V 0 ( π a Δ a + p ϕ Δ ϕ ) = O ( ϵ 4 ) . (194) \mathcal{F}_{\mathrm{can}}=\mathcal{F}_{0}-V_{0}(\pi_{a}\Delta a+p_{\phi}\Delta\phi)=O(\epsilon^{4}). \tag*{(194)} F can = F 0 − V 0 ( π a Δ a + p ϕ Δ ϕ ) = O ( ϵ 4 ) . (194)
Indeed, K \mathcal{K} K is homogeneous of degree one in ( π a , p ϕ ) (\pi_{a},p_{\phi}) ( π a , p ϕ ) , so Euler’s identity gives π a ∂ π a K + p ϕ ∂ p ϕ K = K \pi_{a}\partial_{\pi_{a}}\mathcal{K}+p_{\phi}\partial_{p_{\phi}}\mathcal{K}=\mathcal{K} π a ∂ π a K + p ϕ ∂ p ϕ K = K . Thus the displayed endpoint term cancels through quadratic order after completing the homogeneous variables. The transformation is canonical through the order required for the stated minimal cubic extension. This cancellation concerns this particular canonical completion; boundary terms generated by other integrations by parts, and initial-state transformations between different canonical representations, require their own accounting.
It remains to check how the quadratic background redefinitions affect the perturbative expansion. Denote collectively the homogeneous variables by W A W^{A} W A and the perturbative variables by V V V . Since
Δ W A = O ( ϵ 2 ) , (195) \Delta W^{A}=O(\epsilon^{2}), \tag*{(195)} Δ W A = O ( ϵ 2 ) , (195)
and
W A = W ~ A − Δ W A , (196) W^{A}=\widetilde{W}^{A}-\Delta W^{A}, \tag*{(196)} W A = W A − Δ W A , (196)
the background Hamiltonian transforms according to
H ( 0 ) ( W ) = H ( 0 ) ( W ~ ) − Δ W A ∂ H ( 0 ) ∂ W A ∣ W ~ + O ( ϵ 4 ) . (197) H^{(0)}(W)=H^{(0)}(\widetilde{W})-\left.\Delta W^{A}\frac{\partial H^{(0)}}{\partial W^{A}}\right|_{\widetilde{W}}+O(\epsilon^{4}). \tag*{(197)} H ( 0 ) ( W ) = H ( 0 ) ( W ) − Δ W A ∂ W A ∂ H ( 0 ) W + O ( ϵ 4 ) . (197)
The second term in Eq. (B.17) is of quadratic perturbative order and is therefore part of the completed quadratic Hamiltonian. For the second-order Hamiltonian, we find
H ( 2 ) ( W , V ) = H ( 2 ) ( W ~ , V ) + O ( ϵ 4 ) , (198) H^{(2)}(W,V)=H^{(2)}(\widetilde{W},V)+O(\epsilon^{4}), \tag*{(198)} H ( 2 ) ( W , V ) = H ( 2 ) ( W , V ) + O ( ϵ 4 ) , (198)
because Δ W A ∂ A H ( 2 ) = O ( ϵ 4 ) \Delta W^{A}\partial_{A}H^{(2)}=O(\epsilon^{4}) Δ W A ∂ A H ( 2 ) = O ( ϵ 4 ) . Likewise, we find
H ( 3 ) ( W , V ) = H ( 3 ) ( W ~ , V ) + O ( ϵ 5 ) . (199) H^{(3)}(W,V)=H^{(3)}(\widetilde{W},V)+O(\epsilon^{5}). \tag*{(199)} H ( 3 ) ( W , V ) = H ( 3 ) ( W , V ) + O ( ϵ 5 ) . (199)
Therefore the quadratic background shifts required for the canonical completion do not generate any additional cubic contribution.
The only other possible source of such a contribution would be a term of the schematic form Δ W ( 2 ) ∂ W H ( 1 ) \Delta W^{(2)}\partial_{W}H^{(1)} Δ W ( 2 ) ∂ W H ( 1 ) . However, after imposing the spatially flat gauge and solving the first-order scalar and vector constraints, the reduced perturbative Hamiltonian starts at second order,
H p e r t = H ( 2 ) + H ( 3 ) + ⋯ . (200) H_{\mathrm{pert}}=H^{(2)}+H^{(3)}+\cdots. \tag*{(200)} H pert = H ( 2 ) + H ( 3 ) + ⋯ . (200)
Hence no such cubic term is present.
We conclude that, within the minimal nonlinear extension in which ν = a δ ϕ \nu=a\,\delta\phi ν = a δ ϕ is kept linear, the full canonical completion leaves the explicit cubic Hamiltonian unchanged:
H ~ ( 3 ) = H ( 3 ) [ δ ϕ = ν a , δ p ϕ = a ( π ν + κ π a 6 a ν ) ] . (201) \widetilde{H}^{(3)}=H^{(3)}\left[\delta\phi=\frac{\nu}{a},\ \delta p_{\phi}=a\left(\pi_{\nu}+\frac{\kappa\pi_{a}}{6a}\nu\right)\right]. \tag*{(201)} H ( 3 ) = H ( 3 ) [ δ ϕ = a ν , δ p ϕ = a ( π ν + 6 a κ π a ν ) ] . (201)
All homogeneous variables appearing on the right-hand side of Eq. (B.21) are understood as the canonically completed variables, with the tildes suppressed for notational simplicity.
Independent numerical validation and a fixed-window remote-state test
This appendix reports independent implementations of selected background, mode, vertex, and convolution equations. Their purpose is to distinguish algebraic consistency, numerical stability on a specified finite domain, and the physical scope of the retained diagnostic. Agreement of two implementations sharing a Hamiltonian does not independently establish that Hamiltonian or control contributions omitted from Eq. (116) . The original numerical tables are retained: the explicitly specified tests below do not reconstruct every initial-state, averaging, and momentum-window convention underlying them.
Conventions and controlled benchmarks
The independent calculations set G = ℏ = c = 1 G=\hbar=c=1 G = ℏ = c = 1 , with the unreduced Planck mass m P = 1 m_{\mathrm{P}}=1 m P = 1 and κ = 8 π \kappa=8\pi κ = 8 π . The representative inputs are
γ = 0.2375 , λ 2 = 4 3 π γ ℓ P 2 , \gamma=0.2375,\qquad\lambda^2=4\sqrt{3}\pi\gamma\ell_{\mathrm{P}}^2, γ = 0.2375 , λ 2 = 4 3 π γ ℓ P 2 ,
m ⋆ = 3.08 × 10 − 6 m P , ϕ B = − 1.09 m P , (202) m_{\star}=3.08\times10^{-6}m_{\mathrm{P}},\qquad\phi_{\mathrm{B}}=-1.09m_{\mathrm{P}}, \tag*{(202)} m ⋆ = 3.08 × 1 0 − 6 m P , ϕ B = − 1.09 m P , (202)
with a B = 1 a_{\mathrm{B}}=1 a B = 1 and positive ϕ ˙ B \dot{\phi}_{\mathrm{B}} ϕ ˙ B . These two scale choices are kept distinct when comparing rapidly oscillating quantities. On the expanding branch the Hamiltonian-constraint root is used after the early evolution; subsequent constraint residuals are therefore algebraically enforced, rather than an independent conservation test.
The in-in sign and contraction factor are checked with a separate finite-time oscillator, H = ω ( n + 1 / 2 ) + g q 3 H=\omega(n+1/2)+gq^3 H = ω ( n + 1/2 ) + g q 3 , initialized in the free vacuum. The connected cumulant, rather than the raw third moment containing a tadpole, satisfies at first order
κ 3 ( T ) = − g 2 ω 4 [ 1 − cos ( 3 ω T ) ] + O ( g 3 ) . (203) \kappa_3(T)=-\frac{g}{2\omega^4}\left[1-\cos(3\omega T)\right]+O(g^3). \tag*{(203)} κ 3 ( T ) = − 2 ω 4 g [ 1 − cos ( 3 ω T ) ] + O ( g 3 ) . (203)
Finite-Fock-space evolution and coupling extrapolation reproduce this coefficient to approximately 1.3 × 10 − 12 1.3\times10^{-12} 1.3 × 1 0 − 12 relatively in the tested cases. This is a finite-time weak-coupling benchmark, not a statement about the ground state of an unbounded cubic potential.
A second benchmark uses a separate classical GR background with constant ϵ = − H ˙ / H 2 \epsilon=-\dot{H}/H^2 ϵ = − H ˙ / H 2 and analytic Bunch–Davies Hankel modes. With the negative f N L f_{\mathrm{NL}} f NL convention of Eq. (153) , the leading slow-roll equilateral and exact power-law soft-limit references are
f N L e q u i l = − 55 36 ϵ + O ( ϵ 2 ) , f_{\mathrm{NL}}^{\mathrm{equil}}=-\frac{55}{36}\epsilon+O(\epsilon^2), f NL equil = − 36 55 ϵ + O ( ϵ 2 ) ,
f N L s o f t = − 5 ϵ 6 ( 1 − ϵ ) . (204) f_{\mathrm{NL}}^{\mathrm{soft}}=-\frac{5\epsilon}{6(1-\epsilon)}. \tag*{(204)} f NL soft = − 6 ( 1 − ϵ ) 5 ϵ . (204)
The finite-momentum shape agrees with the leading result of [29 ] under a small-ϵ \epsilon ϵ extrapolation at the 10 − 5 10^{-5} 1 0 − 5 level or better for the three tested shapes. At long-to-short momentum ratio 5 × 10 − 4 5\times10^{-4} 5 × 1 0 − 4 and a refined late surface, the soft-limit residual is about 4.5 × 10 − 7 4.5\times10^{-7} 4.5 × 1 0 − 7 in the tested cases. These residuals include finite-momentum and finite-final-time effects. The GR vacuum contour and soft relation are not imposed on the bounce or remote candidate states. An independent symplectic check also supports the minimal canonical completion in Appendix B; it does not establish uniqueness of the full quantum nonlinear transformation. A separate finite-dimensional unitary-transformation test of the cubic oscillator reproduces the connected evolution with consistently transformed initial data; resetting the free Gaussian changes the result. This is a boundary/state-bookkeeping benchmark, not evidence for an unidentified boundary term in the mLQC-I Hamiltonian.
Exact radial separation and oscillatory time quadrature
Expanding the scalar products and inverse squared momenta in Eq. (101) gives a finite separable representation,
V k p q = ∑ ℓ = 1 49 c ℓ ( η ) ∏ j = 1 3 k j 2 e ℓ j X ℓ j , k j ( η ) , (205) \mathcal{V}_{kpq}=\sum_{\ell=1}^{49}c_{\ell}(\eta)\prod_{j=1}^{3}k_j^{2e_{\ell j}}X_{\ell j,k_j}(\eta), \tag*{(205)} V k pq = ℓ = 1 ∑ 49 c ℓ ( η ) j = 1 ∏ 3 k j 2 e ℓ j X ℓ j , k j ( η ) , (205)
where ( k 1 , k 2 , k 3 ) = ( k , p , q ) (k_1,k_2,k_3)=(k,p,q) ( k 1 , k 2 , k 3 ) = ( k , p , q ) , each X X X is one of the mode combinations in (99) , and e ℓ j ∈ { − 1 , 0 , 1 , 2 } e_{\ell j}\in\{-1,0,1,2\} e ℓ j ∈ { − 1 , 0 , 1 , 2 } . The coefficients absorb all background factors. Random-triangle algebra checks of this 49-term expansion agree with the compact vertex to about 2.7 × 10 − 13 2.7\times10^{-13} 2.7 × 1 0 − 13 relatively in the tested sample.
For example, an internal contraction is organized as
∫ p − p + d p F ℓ ( p , η ) ∫ q − ( k , p ) q + ( k , p ) d q G ℓ ( q , η ) , (206) \int_{p_-}^{p_+}dp\,F_{\ell}(p,\eta)\int_{q_-(k,p)}^{q_+(k,p)}dq\,G_{\ell}(q,\eta), \tag*{(206)} ∫ p − p + d p F ℓ ( p , η ) ∫ q − ( k , p ) q + ( k , p ) d q G ℓ ( q , η ) , (206)
with the final external modes and physical p q pq pq measure included in the factors. Cumulative one-dimensional primitives retain the actual moving triangle limits. The algebraic separation is exact; interpolation and quadrature of its factors are not. It also does not turn an integral of the absolute bispectrum into a product of absolute vertex integrals.
The numerical implementation evolves r k = u k / a r_k=u_k/a r k = u k / a and the corresponding rescaled momentum and shift combinations. As a separate reduction check, writing S k p q = B V ( k , p , q ) / a f 3 S_{kpq}=B_{\mathcal{V}}(k,p,q)/a_{\mathrm{f}}^3 S k pq = B V ( k , p , q ) / a f 3 gives
δ B = − C R H f 2 π 2 κ ∣ ϕ f ∣ ∣ r k , f ∣ 2 ∫ d p d q p q S k p q , \delta_{\mathrm{B}}=-\frac{C_{\mathrm{R}}H_{\mathrm{f}}}{2\pi^2\kappa\lvert\phi_{\mathrm{f}}\rvert\lvert r_{k,\mathrm{f}}\rvert^2}\int dp\,dq\,pq\,S_{kpq}, δ B = − 2 π 2 κ ∣ ϕ f ∣ ∣ r k , f ∣ 2 C R H f ∫ d p d q pq S k pq ,
δ W = C R 2 H f 2 2 π 2 κ ϕ ˙ f 2 ∣ r k , f ∣ 2 ∫ d p d q p q ∣ r p , f ∣ 2 ∣ r q , f ∣ 2 . (207) \delta_{\mathrm{W}}=\frac{C_{\mathrm{R}}^2H_{\mathrm{f}}^2}{2\pi^2\kappa\dot{\phi}_{\mathrm{f}}^2\lvert r_{k,\mathrm{f}}\rvert^2}\int dp\,dq\,pq\,\lvert r_{p,\mathrm{f}}\rvert^2\lvert r_{q,\mathrm{f}}\rvert^2. \tag*{(207)} δ W = 2 π 2 κ ϕ ˙ f 2 ∣ r k , f ∣ 2 C R 2 H f 2 ∫ d p d q pq ∣ r p , f ∣ 2 ∣ r q , f ∣ 2 . (207)
Both integrals use the same finite triangle domain. Independent reductions of stored kernels check the measure and prefactors. Direct triangle and cumulative-radial implementations agree on local square and moving-boundary panels and on a common full-domain coarse lattice. Agreement on the same undersampled lattice is not a continuum-convergence result.
Late and time-dependent modes are represented locally by complex polynomial envelopes multiplying explicit sine and cosine carriers. Separate momentum nodes validate these representations. On the final-time mode test, the worst sampled panel RMS-relative residual for degree-ten envelopes is about 4.9 × 10 − 10 4.9 \times10^{-10} 4.9 × 1 0 − 10 . Fine radial grids are evaluated from these interpolants; they are not newly propagated modes at every fine-grid momentum. Composite Simpson rules, with a final three-eighths block when required, preserve the aligned triangle endpoints. Their spatial error is tested separately.
Ordinary time refinement remains inefficient on the rapidly oscillating post-bounce interval. For the control window below, candidate time carriers are derived from the domain vertices through ∣ ± k ± p v ± q v ∣ \lvert\pm k \pm p_v \pm q_v \rvert ∣ ± k ± p v ± q v ∣ . In Planck units they are
{ 0 , 0.1 , 0.3 , 0.5 , 0.6 , 0.7 , 1.5 , 1.8 , 2.1 , 2.4 } . (208) \{0, 0.1, 0.3, 0.5, 0.6, 0.7, 1.5, 1.8, 2.1, 2.4\}. \tag*{(208)} { 0 , 0.1 , 0.3 , 0.5 , 0.6 , 0.7 , 1.5 , 1.8 , 2.1 , 2.4 } . (208)
Their sufficiency for the radially integrated kernel is a numerical hypothesis, not an exact single-harmonic identity. On each local interval, slow polynomial envelopes in N N N are fitted to these carriers in η ( N ) \eta(N) η ( N ) . The time contraction keeps the Jacobian explicitly,
∫ g ( N ) d N = ∫ a H g [ η ( N ) ] d η . (209) \int g(N)\,\mathrm{d}N = \int aH\,g[\eta(N)]\,\mathrm{d}\eta. \tag*{(209)} ∫ g ( N ) d N = ∫ a H g [ η ( N )] d η . (209)
Sine/cosine-weighted quadrature integrates each carrier; the zero-frequency term is integrated directly. Synthetic weighted-quadrature checks agree with direct high-order integration to about 7.2 × 10 − 14 7.2 \times10^{-14} 7.2 × 1 0 − 14 absolute. Physical short-interval checks on two pilot panels also agree, with suitable envelope degrees, at approximately 10 − 15 10^{-15} 1 0 − 15 absolute or better. Broad fits and rank-deficient higher-degree fits that fail validation are rejected.
The critical N ∈ [ 2 , 6 ] N \in[2,6] N ∈ [ 2 , 6 ] interval is divided into sixteen panels of width 0.25. Each uses 129 Chebyshev training times and 32 separate validation times, giving 2561 unique physical time nodes. Acceptance requires full column rank, condition number below 10 10 10^{10} 1 0 10 , sampled relative RMS below 10 − 4 10^{-4} 1 0 − 4 , maximum sampled absolute residual times panel width below 10 − 10 10^{-10} 1 0 − 10 , and stability against an independently eligible higher degree. Ordinary polynomial quadrature is retained on the four N ∈ [ 2 , 3 ] N \in[2,3] N ∈ [ 2 , 3 ] panels where it already resolves the oscillations; the twelve N ∈ [ 3 , 6 ] N \in[3,6] N ∈ [ 3 , 6 ] panels use phase-aware quadrature. All sixteen pass these gates and a final test on 128 additional time nodes not used to fit or select the models.
For the t i = − 25 t P t_i=-25t_{\mathrm P} t i = − 25 t P candidate, the N ∈ [ 2 , 6 ] N \in[2,6] N ∈ [ 2 , 6 ] contribution to δ B \delta_B δ B is approximately 2.585 × 10 − 8 2.585 \times10^{-8} 2.585 × 1 0 − 8 . The sums of maximum-sampled-residual-times-width indicators from the selection and final tests are about 6.7 × 10 − 11 6.7 \times10^{-11} 6.7 × 1 0 − 11 and 5.4 × 10 − 11 5.4 \times10^{-11} 5.4 × 1 0 − 11 ; the summed higher-degree integral changes are about 8.0 × 10 − 13 8.0 \times10^{-13} 8.0 × 1 0 − 13 . Matched radial-grid checks give a separate sampled indicator of approximately 5.2 × 10 − 10 5.2 \times10^{-10} 5.2 × 1 0 − 10 . Tightening the mode tolerance from 10 − 12 10^{-12} 1 0 − 12 to 2 × 10 − 13 2 \times10^{-13} 2 × 1 0 − 13 on the final test nodes changes physical radial derivatives by at most about 1.3 × 10 − 16 1.3 \times10^{-16} 1.3 × 1 0 − 16 . None of these sampled indicators or internal quadrature estimates is a certified total physical error bound.
Table VI. Finite-start robustness of the independent remote candidate on Eq. (C.9). Percentage differences use the matched t i = − 25 t P t_i=-25t_{\mathrm P} t i = − 25 t P estimate as denominator. They describe differences of numerical estimates, not certified total accuracies.
Conditional signed result and finite initial-time robustness
The fixed-window remote candidate calculation uses
k = 0.3 m P , 0.05 ≤ p m P , q m P ≤ 1.05 , N f = 18 , (210) k=0.3m_{\mathrm P}, \qquad0.05 \le\frac{p}{m_{\mathrm P}},\frac{q}{m_{\mathrm P}} \le1.05,\qquad N_f=18, \tag*{(210)} k = 0.3 m P , 0.05 ≤ m P p , m P q ≤ 1.05 , N f = 18 , (210)
with the triangle inequalities imposed throughout. This is neither the scale-adapted window of Eq. (5.11) nor the enlarged five-point scan. Combining the phase-aware N ∈ [ 2 , 6 ] N \in[2,6] N ∈ [ 2 , 6 ] integral with the carrier history and small later-time tail gives, for the specified t i = − 25 t P t_i=-25t_{\mathrm P} t i = − 25 t P candidate,
δ B ≃ 3.4485 × 10 − 4 , δ W ≃ 9.5469 × 10 − 10 , Ξ ≃ 3.45 × 10 − 4 . (211) \begin{aligned}
\delta_B &\simeq3.4485 \times10^{-4},\\
\delta_W &\simeq9.5469 \times10^{-10},\qquad\Xi\simeq3.45 \times10^{-4}. \tag*{(211)}
\end{aligned} δ B δ W ≃ 3.4485 × 1 0 − 4 , ≃ 9.5469 × 1 0 − 10 , Ξ ≃ 3.45 × 1 0 − 4 . (211)
Only the rounded Ξ \Xi Ξ should be interpreted as the present conditional numerical estimate. The dominant contribution comes from the contracting history, particularly t / t P ∈ [ − 7 , − 4 ] t/t_{\mathrm P} \in[-7,-4] t / t P ∈ [ − 7 , − 4 ] . Its degree-ten to degree-twelve envelope comparison is stable, while a separate radial refinement from 80001 to 160001 points changes that dominant panel by about 1.9 × 10 − 9 1.9 \times10^{-9} 1.9 × 1 0 − 9 absolutely. The smaller N ∈ [ 6 , 18 ] N \in[6,18] N ∈ [ 6 , 18 ] tail retains its separate temporal-resolution qualification; it is not newly phase-validated by the N ∈ [ 2 , 6 ] N \in[2,6] N ∈ [ 2 , 6 ] test. A matched final-surface comparison is reported in Appendix C4; window refinements, background errors, and interpolation amplification under cancellation remain separate uncertainties.
Fresh starts at t i / t P = − 20 , − 25 , − 30 t_i/t_{\mathrm P}=-20,-25,-30 t i / t P = − 20 , − 25 , − 30 provide a full signed-convolution comparison, not merely a comparison of late powers or equilateral bispectra. To isolate the starting prescription, the early radial grid is matched at 40001 points in all three runs; the critical later radial grid is 20001 points with separate 40001-point controls. This matched reference therefore differs slightly from the higher-resolution early integral used in Eq. (C.10). It is not a new physical shift.
All three values round to Ξ ≃ 3.45 × 10 − 4 \Xi\simeq3.45 \times10^{-4} Ξ ≃ 3.45 × 1 0 − 4 . Both new runs also pass their independent final time-node tests. Linear reuse of the stored histories is checked through the canonical identity u k n e w = α k u k r e f + β k u k r e f ∗ u_k^{\mathrm{new}}=\alpha_k u_k^{\mathrm{ref}}+\beta_k u_k^{\mathrm{ref}*} u k new = α k u k ref + β k u k ref ∗ at a common time, with fresh direct forward evolutions providing a separate comparison. The maximum final-mode discrepancy divided by the mode RMS is below 4.1 × 10 − 10 4.1 \times10^{-10} 4.1 × 1 0 − 10 in these tests. On the dominant early panel, replacing the transported final modes with direct ones changes the integral by less than 1.3 × 10 − 13 1.3 \times10^{-13} 1.3 × 1 0 − 13 . Refining its matched radial grid from 40001 to 80001 points changes the state-to-reference differences by at most about 5.1 × 10 − 12 5.1 \times10^{-12} 5.1 × 1 0 − 12 , much less than the observed starting-prescription shifts.
Each fresh run imposes zero initial connected three-point function at its own surface. Thus both free-state data and nonlinear initial conditions vary. A separate test holds the free reference state fixed and changes only the lower interaction limit with a fresh zero primitive; its shifts from the − 25 t P -25t_{\mathrm{P}} − 25 t P reference are approximately − 1.106 × 10 − 8 -1.106 \times10^{-8} − 1.106 × 1 0 − 8 and + 1.106 × 10 − 8 +1.106 \times10^{-8} + 1.106 × 1 0 − 8 for the − 20 t P -20t_{\mathrm{P}} − 20 t P and − 30 t P -30t_{\mathrm{P}} − 30 t P cutoffs. These effects partially compensate free-state changes in one of the fresh initializations. Consistent transport of one interacting state would retain its accumulated primitive and is a different operation.
The results support a small retained signed correction and finite-start robustness for the specified candidate and domain. They do not establish an infinite-past state, a uniform external-momentum or window bound, the complete same-order two-point correction, an error bound on discarded conversion terms, or a validated primordial-to-CMB projection. In particular, Eq. (C.10) is not a replacement for any entry of Table II .
Table II. Magnitude of the signed correlation-function correction on the common scale-adapted numerical window in Eq. (5.11). The external momentum is expressed in units of the bounce-curvature scale k B k_B k B .
Dependence on the final evaluation surface
A separate matched calculation holds the candidate state at t i = − 25 t P t_i=-25t_{\mathrm{P}} t i = − 25 t P , the external momentum k = 0.3 m P k=0.3m_{\mathrm{P}} k = 0.3 m P , and the control window in Eq. (C.9) fixed, and varies only the final surface. The interaction history is integrated through each final surface without resetting its accumulated primitive. The early, critical later-time, and Wick radial grids contain 40001, 20001, and 160001 points, respectively. For N F = 18 N_F=18 N F = 18 , 19, 20 the retained diagnostic is, respectively,
Ξ ≃ ( 3.45 , 3.47 , 3.50 ) × 10 − 4 . (212) \Xi\simeq(3.45,\,3.47,\,3.50)\times10^{-4}. \tag*{(212)} Ξ ≃ ( 3.45 , 3.47 , 3.50 ) × 1 0 − 4 . (212)
All sixteen critical time panels and their independent unused time-node checks pass at each final surface. These are sampled numerical checks, not certified bounds on the full integration error.
The matched unrounded estimates increase by 0.734% and 1.479% relative to N F = 18 N_F=18 N F = 18 . This drift closely tracks the local conversion coefficient: C R = 0.00363282 , 0.00365949 , 0.00368655 C_R=0.00363282, 0.00365949, 0.00368655 C R = 0.00363282 , 0.00365949 , 0.00368655 on the three surfaces. In contrast, the leading external curvature power changes by only about 8.60 × 10 − 6 8.60\times10^{-6} 8.60 × 1 0 − 6 fractionally from N F = 18 N_F=18 N F = 18 to 20. The ratio δ B / C R \delta_B/C_R δ B / C R changes by about 7.57 × 10 − 7 7.57\times10^{-7} 7.57 × 1 0 − 7 fractionally. This ratio diagnoses the scaling; it does not replace or redefine Ξ \Xi Ξ .
Tightening the mode-solver tolerance from 10 − 12 10^{-12} 1 0 − 12 to 2 × 10 − 13 2\times10^{-13} 2 × 1 0 − 13 changes the dominant early-panel contribution at N F = 20 N_F=20 N F = 20 by only about 5.3 × 10 − 14 5.3\times10^{-14} 5.3 × 1 0 − 14 .
The retained correction therefore remains small in this test, but is not strictly independent of the late evaluation surface even though the linear curvature power is nearly frozen. Because the other same-order contributions and omitted conversion terms have not been computed, this drift cannot by itself establish the evolution of the complete physical two-point function. The fixed-N F = 18 N_F=18 N F = 18 conditional estimate above remains applicable on its stated surface; a stronger claim of late-time invariance is not supported by this comparison.
Details of the RBF interpolation
For completeness, we summarize the radial-basis interpolation used in Sec. VI.E. For each state s = B , R s=B,R s = B , R we define the N n o d e = 148 N_{\mathrm{node}}=148 N node = 148 nodes
y j = ( x j , U j , V j ) , x j = ln ( k t , j δ m P ) , (213) \mathbf{y}_j=(x_j,U_j,V_j),\qquad x_j=\ln\left(\frac{k_{t,j}}{\delta m_{\mathrm{P}}}\right), \tag*{(213)} y j = ( x j , U j , V j ) , x j = ln ( δ m P k t , j ) , (213)
with U j U_j U j and V j V_j V j defined in Eq. (160) . The fitted data are
d A , j = ln S f , j , d φ , j = Δ φ j , (214) d_{A,j}=\ln S_{f,j},\qquad d_{\varphi,j}=\Delta\varphi_j, \tag*{(214)} d A , j = ln S f , j , d φ , j = Δ φ j , (214)
These are the logarithmic amplitude ratio and relative phase shift used as the two interpolation targets, respectively.
The unconstrained radial-basis functions in Eq. (163) are reconstructed from
\begin{pmatrix}
K+\lambda_X^{(s)}I & P\\
P^T & 0
\end{pmatrix}
\begin{pmatrix}
\mathbf{w}_X^{(s)}\\
c_X^{(s)}
\end{pmatrix}
=
\begin{pmatrix}
\mathbf{d}_X^{(s)}\\
0
\end{pmatrix}, \tag*{(215)}\end{pmatrix}
where
K i j = Φ ( ∣ y i − y j ∣ ) , Φ ( ρ ) = ρ 2 ln ρ , Φ ( 0 ) = 0 , (216) K_{ij}=\Phi(\lvert\mathbf{y}_i-\mathbf{y}_j\rvert),\qquad\Phi(\rho)=\rho^2\ln\rho,\qquad\Phi(0)=0, \tag*{(216)} K ij = Φ (∣ y i − y j ∣) , Φ ( ρ ) = ρ 2 ln ρ , Φ ( 0 ) = 0 , (216)
and
P j = ( 1 , x j , U j , V j ) . (217) P_j=(1,x_j,U_j,V_j). \tag*{(217)} P j = ( 1 , x j , U j , V j ) . (217)
Thus, the polynomial completion is linear,
p X ( s ) ( y ) = c X 0 ( s ) + c X x ( s ) x + c X U ( s ) U + c X V ( s ) V , X = A , φ . (218) p_X^{(s)}(\mathbf{y})=c_{X0}^{(s)}+c_{Xx}^{(s)}x+c_{XU}^{(s)}U+c_{XV}^{(s)}V,\qquad X=A,\varphi. \tag*{(218)} p X ( s ) ( y ) = c X 0 ( s ) + c X x ( s ) x + c X U ( s ) U + c X V ( s ) V , X = A , φ . (218)
The smoothing parameters and polynomial coefficients are given in Table VII .
Table VII. Smoothing parameters and polynomial-completion coefficients for the RBF representation.
The equilateral normalization is imposed by
ln S f ( s ) ( x , U , V ) = A ~ ( s ) ( x , U , V ) − A ~ ( s ) ( x , 0 , 0 ) , (219) \ln S_f^{(s)}(x,U,V)=\widetilde{A}^{(s)}(x,U,V)-\widetilde{A}^{(s)}(x,0,0), \tag*{(219)} ln S f ( s ) ( x , U , V ) = A ( s ) ( x , U , V ) − A ( s ) ( x , 0 , 0 ) , (219)
Δ φ ( s ) ( x , U , V ) = Φ ~ ( s ) ( x , U , V ) − Φ ~ ( s ) ( x , 0 , 0 ) . (220) \Delta\varphi^{(s)}(x,U,V)=\widetilde{\Phi}^{(s)}(x,U,V)-\widetilde{\Phi}^{(s)}(x,0,0). \tag*{(220)} Δ φ ( s ) ( x , U , V ) = Φ ( s ) ( x , U , V ) − Φ ( s ) ( x , 0 , 0 ) . (220)
The node coordinates and fitted targets are those obtained from the 37 triangle shapes at each of the four reference values of k t k_t k t described in Sec. VI.E. Equations (D.1)–(D.8), together with Table VII , summarize the RBF construction used in the numerical analysis.
References