Clocking the End of Cosmic Inflation

Pierre Auclair    Baptiste Blachier    and Christophe Ringeval
(June 20, 2024)
Abstract

Making observable predictions for cosmic inflation requires determining when the wavenumbers of astrophysical interest today exited the Hubble radius during the inflationary epoch. These instants are commonly evaluated using the slow-roll approximation and measured in e-folds ΔN=NNendΔ𝑁𝑁subscript𝑁end\Delta N=N-N_{\mathrm{end}}roman_Δ italic_N = italic_N - italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT, in reference to the e-fold Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT at which inflation ended. Slow roll being necessarily violated towards the end of inflation, both the approximated trajectory and Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT are determined at, typically, one or two e-folds precision. Up to now, such an uncertainty has been innocuous, but this will no longer be the case with the forthcoming cosmological measurements. In this work, we introduce a new and simple analytical method, on top of the usual slow-roll approximation, that reduces uncertainties on ΔNΔ𝑁\Delta Nroman_Δ italic_N to less than a tenth of an e-fold.

1 Introduction

Cosmic Inflation is a phase of accelerated expansion of the primordial universe which addresses various puzzles of the Big-Bang model as, for instance, the so-called horizon problem and the smallness of the spatial curvature today [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]. Inflation also provides a convincing and simple physical explanation for the origin of cosmic structures: they are seeded by vacuum quantum fluctuations of both the metric and a yet unknown scalar degree of freedom [13].

In its simplest incarnation, both the accelerated expansion and the quantum fluctuations are the outcome of a self-gravitating scalar field ϕitalic-ϕ\phiitalic_ϕ, named the inflaton, slowly rolling down its potential energy V(ϕ)𝑉italic-ϕV(\phi)italic_V ( italic_ϕ ). This class of scenarios is a populated landscape counting hundreds of models, all of them making definite predictions which can be confronted by cosmological observations [14, 15, 16]. As of today, 40%percent4040\%40 % of the proposed scenarios in this class have been ruled-out by Cosmic Microwave Background (CMB) and Large Scale Structure measurements [17, 18, 19]. Still, 60%percent6060\%60 % of the remaining models are compatible with the data. With the deployment of ground-based CMB-S4 polarization telescopes [20, 21, 22], the soon-to-be released Euclid satellite data [23, 24], unprecedented galaxy surveys [25], and the search for B𝐵Bitalic_B-mode polarization from space by the LiteBIRD satellite [26], one should reasonably expect many of the remaining models to be disambiguated and tested. This necessitates, however, that theoretical predictions are made at the required accuracy [27, 28].

In a brute-force manner, one can simply solve the field and gravitational evolution numerically. However, the underlying gravity theory is General Relativity, and, as of today, only parts of the inflationary evolution can be solved without approximation [29, 30, 31, 32, 33, 34, 35, 36]. Considering the regime in which the quantum fluctuations do not dominate the dynamics [37, 38, 39, 40, 41, 42], one can alternatively solve for linear and non-linear perturbations numerically, around a homogeneous background, without any other approximations [43, 44, 45, 46, 47, 48, 49, 50, 51]. These methods are accurate as long as the gravitational effects remain small, but they are computationally too much demanding when dealing with hundreds of different models [52, 53].

There exists, however, a model-free and perturbative treatment for the single-field scenarios in which the assumption of “slow roll” can be made. This approach, initiated in Ref. [1] for the tensor modes, has been extended to scalar perturbations in Refs. [54, 55] and generalized to higher orders in [56, 57, 58, 59, 60, 61, 62, 63]. It has also found applications out of the original context and can be extended to other classes of inflationary models [64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85]. In modern terminology, the slow-roll approximation introduces the Hubble-flow functions defined by [59]

ϵi+1(N)dln|ϵi|dN,ϵ0(N)=MPlH,formulae-sequencesubscriptitalic-ϵ𝑖1𝑁derivative𝑁subscriptitalic-ϵ𝑖subscriptitalic-ϵ0𝑁subscript𝑀Pl𝐻\epsilon_{i+1}(N)\equiv\derivative{\ln\left|\epsilon_{i}\right|}{N}\,,\quad% \epsilon_{0}(N)=\dfrac{M_{\scriptscriptstyle{\mathrm{Pl}}}}{H}\,,italic_ϵ start_POSTSUBSCRIPT italic_i + 1 end_POSTSUBSCRIPT ( italic_N ) ≡ divide start_ARG roman_d start_ARG roman_ln | italic_ϵ start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT | end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG , italic_ϵ start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ( italic_N ) = divide start_ARG italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT end_ARG start_ARG italic_H end_ARG , (1.1)

where H(N)𝐻𝑁H(N)italic_H ( italic_N ) denotes the Hubble parameter during inflation and N=lna𝑁𝑎N=\ln aitalic_N = roman_ln italic_a is the number of e-folds, a𝑎aitalic_a being the Friedmann-Lemaître-Robertson-Walker (FLRW) scale factor. For a quasi-de Sitter accelerated expansion, H(N)𝐻𝑁H(N)italic_H ( italic_N ) is nearly constant and all the Hubble-flow functions are expected to be small. It is therefore possible to solve the linearized Einstein’s equations for both the tensor and scalar perturbations by performing a consistent expansion in terms of these ϵisubscriptitalic-ϵ𝑖\epsilon_{i}italic_ϵ start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT functions. Analytical solutions have currently been derived up to third order [63] and they allow us to calculate the primordial power spectra of the comoving curvature perturbation ζ𝜁\zetaitalic_ζ and of the primordial gravitational waves hμνsubscript𝜇𝜈h_{\mu\nu}italic_h start_POSTSUBSCRIPT italic_μ italic_ν end_POSTSUBSCRIPT. For instance, keeping only the first order terms, one gets, for the power spectrum of the curvature perturbations,

𝒫ζ(k)=H28π2MPl2ϵ1[12(C+1)ϵ1Cϵ2(2ϵ1+ϵ2)ln(kk)+],subscript𝒫𝜁𝑘superscriptsubscript𝐻28superscript𝜋2superscriptsubscript𝑀Pl2subscriptitalic-ϵ1delimited-[]12𝐶1subscriptitalic-ϵ1𝐶subscriptitalic-ϵ22subscriptitalic-ϵ1subscriptitalic-ϵ2𝑘subscript𝑘\mathcal{P}_{\zeta}(k)=\dfrac{H_{*}^{2}}{8\pi^{2}M_{\scriptscriptstyle{\mathrm% {Pl}}}^{2}\epsilon_{1*}}\left[1-2(C+1)\epsilon_{1*}-C\epsilon_{2*}-\left(2% \epsilon_{1*}+\epsilon_{2*}\right)\ln\left(\dfrac{k}{k_{*}}\right)+\dots\right],caligraphic_P start_POSTSUBSCRIPT italic_ζ end_POSTSUBSCRIPT ( italic_k ) = divide start_ARG italic_H start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 8 italic_π start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_ϵ start_POSTSUBSCRIPT 1 ∗ end_POSTSUBSCRIPT end_ARG [ 1 - 2 ( italic_C + 1 ) italic_ϵ start_POSTSUBSCRIPT 1 ∗ end_POSTSUBSCRIPT - italic_C italic_ϵ start_POSTSUBSCRIPT 2 ∗ end_POSTSUBSCRIPT - ( 2 italic_ϵ start_POSTSUBSCRIPT 1 ∗ end_POSTSUBSCRIPT + italic_ϵ start_POSTSUBSCRIPT 2 ∗ end_POSTSUBSCRIPT ) roman_ln ( divide start_ARG italic_k end_ARG start_ARG italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT end_ARG ) + … ] , (1.2)

where the constant CγE+ln(2)20.7296𝐶subscript𝛾E22similar-to-or-equals0.7296C\equiv\gamma_{\scriptscriptstyle{\mathrm{E}}}+\ln(2)-2\simeq-0.7296italic_C ≡ italic_γ start_POSTSUBSCRIPT roman_E end_POSTSUBSCRIPT + roman_ln ( start_ARG 2 end_ARG ) - 2 ≃ - 0.7296 and ksubscript𝑘k_{*}italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT is a wavenumber around which the expansion is made (an observer choice). For the wavenumbers probed by the Cosmic Microwave Background anisotropies, one usually takes k/a0=0.05Mpc1subscript𝑘subscript𝑎00.05superscriptMpc1k_{*}/a_{\scriptscriptstyle{0}}=0.05\,\mathrm{Mpc}^{-1}italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT / italic_a start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT = 0.05 roman_Mpc start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT to be in the middle of the observable range of modes. All the other “starred” quantities in Eq. 1.2 refer to the Hubble-flow functions evaluated at a given e-fold number Nsubscript𝑁N_{*}italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT, i.e., ϵi=ϵi(N)subscriptitalic-ϵ𝑖subscriptitalic-ϵ𝑖subscript𝑁\epsilon_{i*}=\epsilon_{i}(N_{*})italic_ϵ start_POSTSUBSCRIPT italic_i ∗ end_POSTSUBSCRIPT = italic_ϵ start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) and H=H(N)subscript𝐻𝐻subscript𝑁H_{*}=H(N_{*})italic_H start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT = italic_H ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ). This e-fold number is the time at which the physical pivot wavenumber k/asubscript𝑘𝑎k_{*}/aitalic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT / italic_a exited the Hubble radius during inflation, namely the solution of111For practical reasons, Nsubscript𝑁N_{*}italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT is usually defined in terms of the conformal time η𝜂\etaitalic_η by kη(N)=1subscript𝑘𝜂subscript𝑁1k_{*}\eta(N_{*})=-1italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT italic_η ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) = - 1, which coincides with Hubble radius crossing k=a(N)H(N)subscript𝑘𝑎subscript𝑁𝐻subscript𝑁k_{*}=a(N_{*})H(N_{*})italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT = italic_a ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) italic_H ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) at leading order in the Hubble-flow functions. For the present discussion, these differences will not play a role, but they are important when considering higher order terms [63].

ka(N)H(N).similar-to-or-equalssubscript𝑘𝑎subscript𝑁𝐻subscript𝑁k_{*}\simeq a(N_{*})H(N_{*}).italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ≃ italic_a ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) italic_H ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) . (1.3)

As a result, even if the accuracy at which Eq. 1.2 is derived is under control, another source of uncertainties in making observable predictions comes from our ability to determine a precise value for Nsubscript𝑁N_{*}italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT, and, as we will see, for ΔNNNendΔsubscript𝑁subscript𝑁subscript𝑁end\Delta N_{*}\equiv N_{*}-N_{\mathrm{end}}roman_Δ italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ≡ italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT - italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. This is the main focus of this paper.

Before entering into details, let us further express Eq. 1.3 in terms of observable quantities. The physical pivot wavenumber is measured today, for a scale factor given by a0subscript𝑎0a_{\scriptscriptstyle{0}}italic_a start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT, in terms of which Eq. 1.3 reads

ka0=(1+zend)1a(N)a(Nend)H(N).subscript𝑘subscript𝑎0superscript1subscript𝑧end1𝑎subscript𝑁𝑎subscript𝑁end𝐻subscript𝑁\dfrac{k_{*}}{a_{\scriptscriptstyle{0}}}=\left(1+z_{\mathrm{end}}\right)^{-1}% \dfrac{a(N_{*})}{a(N_{\mathrm{end}})}H(N_{*}).divide start_ARG italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT end_ARG start_ARG italic_a start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT end_ARG = ( 1 + italic_z start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) start_POSTSUPERSCRIPT - 1 end_POSTSUPERSCRIPT divide start_ARG italic_a ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) end_ARG start_ARG italic_a ( italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) end_ARG italic_H ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) . (1.4)

We have made explicit zend=a0/aend1subscript𝑧endsubscript𝑎0subscript𝑎end1z_{\mathrm{end}}=a_{\scriptscriptstyle{0}}/a_{\mathrm{end}}-1italic_z start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = italic_a start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT / italic_a start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT - 1, the redshift at which inflation ended. It depends on the universe history after inflation and, in particular, it is sensitive to the so-called reheating era. Following Refs. [52, 86], one can conveniently absorb all the kinematic effects associated with this era into the reheating parameter Rradsubscript𝑅radR_{\mathrm{rad}}italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT defined by

Rradaendareh(ρendρreh)1/4.subscript𝑅radsubscript𝑎endsubscript𝑎rehsuperscriptsubscript𝜌endsubscript𝜌reh14R_{\mathrm{rad}}\equiv\dfrac{a_{\mathrm{end}}}{a_{\mathrm{reh}}}\left(\dfrac{% \rho_{\mathrm{end}}}{\rho_{\mathrm{reh}}}\right)^{1/4}.italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT ≡ divide start_ARG italic_a start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_ARG start_ARG italic_a start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT end_ARG ( divide start_ARG italic_ρ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_ARG start_ARG italic_ρ start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT end_ARG ) start_POSTSUPERSCRIPT 1 / 4 end_POSTSUPERSCRIPT . (1.5)

Here ρendsubscript𝜌end\rho_{\mathrm{end}}italic_ρ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT and ρrehsubscript𝜌reh\rho_{\mathrm{reh}}italic_ρ start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT stand for the energy density of the universe at the end of inflation and at beginning of the radiation era (the end reheating), respectively. In terms of Rradsubscript𝑅radR_{\mathrm{rad}}italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT, one has

1+zend=1Rrad(ρend3𝒬rehΩradH02)1/4,1subscript𝑧end1subscript𝑅radsuperscriptsubscript𝜌end3subscript𝒬rehsubscriptΩradsuperscriptsubscript𝐻02141+z_{\mathrm{end}}=\dfrac{1}{R_{\mathrm{rad}}}\left(\dfrac{\rho_{\mathrm{end}}% }{3\mathcal{Q}_{\mathrm{reh}}\Omega_{\mathrm{rad}}H_{\scriptscriptstyle{0}}^{2% }}\right)^{1/4},1 + italic_z start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = divide start_ARG 1 end_ARG start_ARG italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT end_ARG ( divide start_ARG italic_ρ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_ARG start_ARG 3 caligraphic_Q start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT roman_Ω start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT italic_H start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG ) start_POSTSUPERSCRIPT 1 / 4 end_POSTSUPERSCRIPT , (1.6)

where 𝒬rehq04/3greh/(qreh4/3g0)subscript𝒬rehsuperscriptsubscript𝑞043subscript𝑔rehsuperscriptsubscript𝑞reh43subscript𝑔0\mathcal{Q}_{\mathrm{reh}}\equiv q_{\scriptscriptstyle{0}}^{4/3}g_{\mathrm{reh% }}/(q_{\mathrm{reh}}^{4/3}g_{\scriptscriptstyle{0}})caligraphic_Q start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT ≡ italic_q start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 4 / 3 end_POSTSUPERSCRIPT italic_g start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT / ( italic_q start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 4 / 3 end_POSTSUPERSCRIPT italic_g start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ) is a measure of the change of number of entropic (q𝑞qitalic_q) and energetic (g𝑔gitalic_g) relativistic degrees of freedom between the beginning of the radiation era and today [87]. For instance, one has 𝒬reh0.39similar-to-or-equalssubscript𝒬reh0.39\mathcal{Q}_{\mathrm{reh}}\simeq 0.39caligraphic_Q start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT ≃ 0.39 for the Standard Model [88]. One can further expand Eqs. 1.4 to 1.6 for single-field inflationary models by making use of the Friedmann-Lemaître equations for a self gravitating scalar field ϕitalic-ϕ\phiitalic_ϕ. As shown in Section 2, they allow us to express the Hubble parameter during inflation as

H2=ρ3MPl2=1MPl2V(ϕ)3ϵ1.superscript𝐻2𝜌3superscriptsubscript𝑀Pl21superscriptsubscript𝑀Pl2𝑉italic-ϕ3subscriptitalic-ϵ1H^{2}=\dfrac{\rho}{3M_{\scriptscriptstyle{\mathrm{Pl}}}^{2}}=\dfrac{1}{M_{% \scriptscriptstyle{\mathrm{Pl}}}^{2}}\dfrac{V(\phi)}{3-\epsilon_{1}}\,.italic_H start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT = divide start_ARG italic_ρ end_ARG start_ARG 3 italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG = divide start_ARG 1 end_ARG start_ARG italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG divide start_ARG italic_V ( italic_ϕ ) end_ARG start_ARG 3 - italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_ARG . (1.7)

At the end of inflation, one therefore has

ρend=3Vend3ϵ1end=VendV3V3ϵ1end=3MPl2H2VendV3ϵ13ϵ1end.subscript𝜌end3subscript𝑉end3subscriptitalic-ϵ1endsubscript𝑉endsubscript𝑉3subscript𝑉3subscriptitalic-ϵ1end3superscriptsubscript𝑀Pl2superscriptsubscript𝐻2subscript𝑉endsubscript𝑉3subscriptitalic-ϵ13subscriptitalic-ϵ1end\rho_{\mathrm{end}}=\dfrac{3V_{\mathrm{end}}}{3-\epsilon_{1\mathrm{end}}}=% \dfrac{V_{\mathrm{end}}}{V_{*}}\dfrac{3V_{*}}{3-\epsilon_{1\mathrm{end}}}=3M_{% \scriptscriptstyle{\mathrm{Pl}}}^{2}H_{*}^{2}\dfrac{V_{\mathrm{end}}}{V_{*}}% \dfrac{3-\epsilon_{1*}}{3-\epsilon_{1\mathrm{end}}}\,.italic_ρ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = divide start_ARG 3 italic_V start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_ARG start_ARG 3 - italic_ϵ start_POSTSUBSCRIPT 1 roman_e roman_n roman_d end_POSTSUBSCRIPT end_ARG = divide start_ARG italic_V start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_ARG start_ARG italic_V start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT end_ARG divide start_ARG 3 italic_V start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT end_ARG start_ARG 3 - italic_ϵ start_POSTSUBSCRIPT 1 roman_e roman_n roman_d end_POSTSUBSCRIPT end_ARG = 3 italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_H start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT divide start_ARG italic_V start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_ARG start_ARG italic_V start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT end_ARG divide start_ARG 3 - italic_ϵ start_POSTSUBSCRIPT 1 ∗ end_POSTSUBSCRIPT end_ARG start_ARG 3 - italic_ϵ start_POSTSUBSCRIPT 1 roman_e roman_n roman_d end_POSTSUBSCRIPT end_ARG . (1.8)

Plugging this expression into Eq. 1.6, taking the logarithm of Eq. 1.4, one finally gets

ΔNNNend=lnRrad+N0+14ln[9ϵ1(3ϵ1end)VendV]14ln(8π2P0),Δsubscript𝑁subscript𝑁subscript𝑁endsubscript𝑅radsubscript𝑁0149subscriptitalic-ϵ13subscriptitalic-ϵ1endsubscript𝑉endsubscript𝑉148superscript𝜋2subscript𝑃0\Delta N_{*}\equiv N_{*}-N_{\mathrm{end}}=-\ln R_{\mathrm{rad}}+N_{% \scriptscriptstyle{0}}+\dfrac{1}{4}\ln\left[\dfrac{9}{\epsilon_{1*}\left(3-% \epsilon_{1\mathrm{end}}\right)}\dfrac{V_{\mathrm{end}}}{V_{*}}\right]-\dfrac{% 1}{4}\ln\left(8\pi^{2}P_{\scriptscriptstyle{0}}\right),roman_Δ italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ≡ italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT - italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = - roman_ln italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT + italic_N start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT + divide start_ARG 1 end_ARG start_ARG 4 end_ARG roman_ln [ divide start_ARG 9 end_ARG start_ARG italic_ϵ start_POSTSUBSCRIPT 1 ∗ end_POSTSUBSCRIPT ( 3 - italic_ϵ start_POSTSUBSCRIPT 1 roman_e roman_n roman_d end_POSTSUBSCRIPT ) end_ARG divide start_ARG italic_V start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_ARG start_ARG italic_V start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT end_ARG ] - divide start_ARG 1 end_ARG start_ARG 4 end_ARG roman_ln ( 8 italic_π start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_P start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ) , (1.9)

where N0subscript𝑁0N_{\scriptscriptstyle{0}}italic_N start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT is defined by

N0ln[k/a0(3𝒬rehΩradH02MPl2)1/4]61.5,subscript𝑁0subscript𝑘subscript𝑎0superscript3subscript𝒬rehsubscriptΩradsuperscriptsubscript𝐻02superscriptsubscript𝑀Pl214similar-to-or-equals61.5N_{\scriptscriptstyle{0}}\equiv\ln\left[\dfrac{k_{*}/a_{0}}{\left(3\mathcal{Q}% _{\mathrm{reh}}\Omega_{\mathrm{rad}}H_{\scriptscriptstyle{0}}^{2}M_{% \scriptscriptstyle{\mathrm{Pl}}}^{2}\right)^{1/4}}\right]\simeq-61.5,italic_N start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ≡ roman_ln [ divide start_ARG italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT / italic_a start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT end_ARG start_ARG ( 3 caligraphic_Q start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT roman_Ω start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT italic_H start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ) start_POSTSUPERSCRIPT 1 / 4 end_POSTSUPERSCRIPT end_ARG ] ≃ - 61.5 , (1.10)

the absolute value of which giving the typical number of e-folds of decelerated expansion after inflation. In all practical situations, one has ϵi1much-less-thansubscriptitalic-ϵ𝑖1\epsilon_{i*}\ll 1italic_ϵ start_POSTSUBSCRIPT italic_i ∗ end_POSTSUBSCRIPT ≪ 1 and, for consistency, we have kept only the leading order terms in ϵisubscriptitalic-ϵ𝑖\epsilon_{i*}italic_ϵ start_POSTSUBSCRIPT italic_i ∗ end_POSTSUBSCRIPT while deriving Eq. 1.9. Moreover, in the last term, we have made explicit the quantity P0=H2/(8π2MPl2ϵ1)subscript𝑃0superscriptsubscript𝐻28superscript𝜋2superscriptsubscript𝑀Pl2subscriptitalic-ϵ1P_{\scriptscriptstyle{0}}=H_{*}^{2}/(8\pi^{2}M_{\scriptscriptstyle{\mathrm{Pl}% }}^{2}\epsilon_{1*})italic_P start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT = italic_H start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / ( 8 italic_π start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_ϵ start_POSTSUBSCRIPT 1 ∗ end_POSTSUBSCRIPT ), which is a very well measured observable as P0𝒫ζ(k)=2.097×109similar-to-or-equalssubscript𝑃0subscript𝒫𝜁subscript𝑘2.097superscript109P_{\scriptscriptstyle{0}}\simeq\mathcal{P}_{\zeta}(k_{*})=2.097\times 10^{-9}italic_P start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ≃ caligraphic_P start_POSTSUBSCRIPT italic_ζ end_POSTSUBSCRIPT ( italic_k start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) = 2.097 × 10 start_POSTSUPERSCRIPT - 9 end_POSTSUPERSCRIPT [17].

Let us now explain how to determine the value of Nsubscript𝑁N_{*}italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT under the hypothesis that an inflationary model, given by its potential V(ϕ)𝑉italic-ϕV(\phi)italic_V ( italic_ϕ ), is specified. Any possible reheating history is associated with definite values for lnRradsubscript𝑅rad\ln R_{\mathrm{rad}}roman_ln italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT (and N0subscript𝑁0N_{\scriptscriptstyle{0}}italic_N start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT). For instance, a radiation-like, or an instantaneous reheating, are both associated with lnRrad=0subscript𝑅rad0\ln R_{\mathrm{rad}}=0roman_ln italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT = 0. In this situation, Eq. 1.9 ends up being a simple algebraic equation for Nsubscript𝑁N_{*}italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT provided one has an explicit expression for V=V[ϕ(N)]subscript𝑉𝑉delimited-[]italic-ϕsubscript𝑁V_{*}=V[\phi(N_{*})]italic_V start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT = italic_V [ italic_ϕ ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ) ], Vend=V[ϕ(Nend)]subscript𝑉end𝑉delimited-[]italic-ϕsubscript𝑁endV_{\mathrm{end}}=V[\phi(N_{\mathrm{end}})]italic_V start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = italic_V [ italic_ϕ ( italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) ] and Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. In other words, one must determine the field trajectory ϕ(N)italic-ϕ𝑁\phi(N)italic_ϕ ( italic_N ) to solve Eq. 1.9 at given reheating history. Usually, this cannot be made exactly and one has to resort to an exact numerical integration, or, to some slow roll approximation to evaluate the field trajectory ϕ(N)italic-ϕ𝑁\phi(N)italic_ϕ ( italic_N ). Notice that, it is also possible to interpret Eq. 1.9 as an algebraic equation on ϕ=ϕ(N)subscriptitalic-ϕitalic-ϕsubscript𝑁\phi_{*}=\phi(N_{*})italic_ϕ start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT = italic_ϕ ( italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ), but this still requires determining the field trajectory in order to evaluate N=N(ϕ)subscript𝑁𝑁subscriptitalic-ϕN_{*}=N(\phi_{*})italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT = italic_N ( italic_ϕ start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT ).

The fastest and most practical method used to determine the field trajectory is the slow-roll approximation. As we show in Section 2.2, it induces (1)order1\order{1}( start_ARG 1 end_ARG ) errors, which have been, up to now, not a concern. Indeed, most of the theoretical unknowns in Eq. 1.9 are actually associated with the reheating, namely the values of lnRradsubscript𝑅rad\ln R_{\mathrm{rad}}roman_ln italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT (𝒬rehsubscript𝒬reh\mathcal{Q}_{\mathrm{reh}}caligraphic_Q start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT does not have significant effects provided the number of relativistic degrees of freedom does not take exponentially large values [89]). The actual value of ρrehsubscript𝜌reh\rho_{\mathrm{reh}}italic_ρ start_POSTSUBSCRIPT roman_reh end_POSTSUBSCRIPT is unknown by orders of magnitude and, in principle, it is allowed to vary from a lower bound as small as Big-Bang Nucleosynthesis ρnuc1/4=(MeV)superscriptsubscript𝜌nuc14orderMeV\rho_{\mathrm{nuc}}^{1/4}=\order{\mathrm{MeV}}italic_ρ start_POSTSUBSCRIPT roman_nuc end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 1 / 4 end_POSTSUPERSCRIPT = ( start_ARG roman_MeV end_ARG ) to ρend1/4superscriptsubscript𝜌end14\rho_{\mathrm{end}}^{1/4}italic_ρ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 1 / 4 end_POSTSUPERSCRIPT which can be as large as 1015GeVsuperscript1015GeV10^{15}\,\mathrm{GeV}10 start_POSTSUPERSCRIPT 15 end_POSTSUPERSCRIPT roman_GeV. Under very reasonable assumptions, one can show that lnRrad[46,15]subscript𝑅rad4615\ln R_{\mathrm{rad}}\in[-46,15]roman_ln italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT ∈ [ - 46 , 15 ] (see Ref. [90, 52]).

This justifies why questioning the accuracy at which N(ϕ)𝑁italic-ϕN(\phi)italic_N ( italic_ϕ ) is evaluated was not a concern. However, as shown in Ref. [19], the current cosmological data are now constraining the reheating era and models having exactly the same accelerated inflationary phase but differing only by their reheating histories, i.e., predicting different values of lnRradsubscript𝑅rad\ln R_{\mathrm{rad}}roman_ln italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT, can now be disambiguated. From another point of view, even for inflationary scenarios not specifying the reheating, the current data allow us to determine the favoured values of lnRradsubscript𝑅rad\ln R_{\mathrm{rad}}roman_ln italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT. Any uncertainty in the determination of ΔNΔsubscript𝑁\Delta N_{*}roman_Δ italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT will then bias the constraints on lnRradsubscript𝑅rad\ln R_{\mathrm{rad}}roman_ln italic_R start_POSTSUBSCRIPT roman_rad end_POSTSUBSCRIPT. As such, it is becoming relevant to improve the accuracy at which the function N(ϕ)𝑁italic-ϕN(\phi)italic_N ( italic_ϕ ) can be determined.

The paper is organized as follows. In Section 2, we recap how to obtain the field trajectory within a FLRW metric and detail the slow-roll method commonly used to approximate the solution. In particular, we use a numerical integration to discuss the amplitude and the origin of the uncertainties made by using the slow-roll approximated trajectory instead of the exact one. In Section 3, we present new analytical results and an exact expansion of the trajectory which allow us to propose a “velocity correction” to the traditional slow-roll. We show that such a correction reduces the uncertainties by an order of magnitude. Section 4 is dedicated to the problem of determining the field value ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT at which inflation ends, which is another (small) source of errors on the determination of ρendsubscript𝜌end\rho_{\mathrm{end}}italic_ρ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. We present various analytical approaches to address this issue and test them within various inflationary scenarios. Here as well, we show that our method reduces the uncertainties on ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT by an order of magnitude. Our conclusion are presented in Section 5.

2 Basics on the field trajectory

2.1 Equations of motion

In the following, we assume a minimally coupled single scalar field ϕitalic-ϕ\phiitalic_ϕ within a FLRW metric. The Friedmann-Lemaître and Klein-Gordon equations read

H2superscript𝐻2\displaystyle H^{2}italic_H start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT =13MPl2[12ϕ˙2+V(ϕ)],absent13superscriptsubscript𝑀Pl2delimited-[]12superscript˙italic-ϕ2𝑉italic-ϕ\displaystyle=\dfrac{1}{3M_{\scriptscriptstyle{\mathrm{Pl}}}^{2}}\left[\dfrac{% 1}{2}\dot{\phi}^{2}+V(\phi)\right],= divide start_ARG 1 end_ARG start_ARG 3 italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG [ divide start_ARG 1 end_ARG start_ARG 2 end_ARG over˙ start_ARG italic_ϕ end_ARG start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT + italic_V ( italic_ϕ ) ] , (2.1)
H2+H˙2superscript𝐻2superscript˙𝐻2\displaystyle H^{2}+\dot{H}^{2}italic_H start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT + over˙ start_ARG italic_H end_ARG start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT =16MPl2[2ϕ˙22V(ϕ)],absent16superscriptsubscript𝑀Pl2delimited-[]2superscript˙italic-ϕ22𝑉italic-ϕ\displaystyle=-\dfrac{1}{6M_{\scriptscriptstyle{\mathrm{Pl}}}^{2}}\left[2\dot{% \phi}^{2}-2V(\phi)\right],= - divide start_ARG 1 end_ARG start_ARG 6 italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG [ 2 over˙ start_ARG italic_ϕ end_ARG start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT - 2 italic_V ( italic_ϕ ) ] , (2.2)
ϕ¨¨italic-ϕ\displaystyle\ddot{\phi}over¨ start_ARG italic_ϕ end_ARG +3Hϕ˙+dVdϕ=0,3𝐻˙italic-ϕderivativeitalic-ϕ𝑉0\displaystyle+3H\dot{\phi}+\derivative{V}{\phi}=0,+ 3 italic_H over˙ start_ARG italic_ϕ end_ARG + divide start_ARG roman_d start_ARG italic_V end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG = 0 , (2.3)

where a dot denotes differentiation with respect to the cosmic time t𝑡titalic_t and Ha˙/a𝐻˙𝑎𝑎H\equiv\dot{a}/aitalic_H ≡ over˙ start_ARG italic_a end_ARG / italic_a. In terms of the number of e-fold Nlna𝑁𝑎N\equiv\ln aitalic_N ≡ roman_ln italic_a, these equations can be decoupled. From Eqs. 1.1, 2.1 and 2.2, one has

ϵ1=H˙H2=12MPl2(dϕdN)2,subscriptitalic-ϵ1˙𝐻superscript𝐻212superscriptsubscript𝑀Pl2superscriptderivative𝑁italic-ϕ2\epsilon_{1}=-\dfrac{\dot{H}}{H^{2}}=\dfrac{1}{2M_{\scriptscriptstyle{\mathrm{% Pl}}}^{2}}\left(\derivative{\phi}{N}\right)^{2},italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT = - divide start_ARG over˙ start_ARG italic_H end_ARG end_ARG start_ARG italic_H start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG = divide start_ARG 1 end_ARG start_ARG 2 italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG ( divide start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT , (2.4)

and the first Hubble-flow function ϵ1subscriptitalic-ϵ1\epsilon_{1}italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT measures the kinetic energy of the field when time is counted in e-fold. In order to simplify the notations, let us introduce the “field velocity” in e-fold as

Γ1MPldϕdN=1MPlHϕ˙,Γ1subscript𝑀Plderivative𝑁italic-ϕ1subscript𝑀Pl𝐻˙italic-ϕ\Gamma\equiv\dfrac{1}{M_{\scriptscriptstyle{\mathrm{Pl}}}}\derivative{\phi}{N}% =\dfrac{1}{M_{\scriptscriptstyle{\mathrm{Pl}}}H}\dot{\phi},roman_Γ ≡ divide start_ARG 1 end_ARG start_ARG italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT end_ARG divide start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG = divide start_ARG 1 end_ARG start_ARG italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT italic_H end_ARG over˙ start_ARG italic_ϕ end_ARG , (2.5)

such that ϵ1=Γ2/2subscriptitalic-ϵ1superscriptΓ22\epsilon_{1}=\Gamma^{2}/2italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT = roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / 2. Expressing Eqs. 2.1 and 2.2 in e-fold time, one obtains Eq. 1.7 for the Hubble parameter, which can be finally plugged into Eq. 2.3 to obtain a decoupled equation of motion for ϕ(N)italic-ϕ𝑁\phi(N)italic_ϕ ( italic_N )

13ϵ1d2ϕdN2+dϕdN=MPl2dlnVdϕ.13subscriptitalic-ϵ1derivative𝑁2italic-ϕderivative𝑁italic-ϕsuperscriptsubscript𝑀Pl2derivativeitalic-ϕ𝑉\dfrac{1}{3-\epsilon_{1}}\derivative[2]{\phi}{N}+\derivative{\phi}{N}=-M_{% \scriptscriptstyle{\mathrm{Pl}}}^{2}\derivative{\ln V}{\phi}.divide start_ARG 1 end_ARG start_ARG 3 - italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_ARG divide start_ARG start_DIFFOP SUPERSCRIPTOP start_ARG roman_d end_ARG start_ARG 2 end_ARG end_DIFFOP start_ARG italic_ϕ end_ARG end_ARG start_ARG SUPERSCRIPTOP start_ARG roman_d start_ARG italic_N end_ARG end_ARG start_ARG 2 end_ARG end_ARG + divide start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG = - italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT divide start_ARG roman_d start_ARG roman_ln italic_V end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG . (2.6)

From now on, we will be working in Planck units with MPl=1subscript𝑀Pl1M_{\scriptscriptstyle{\mathrm{Pl}}}=1italic_M start_POSTSUBSCRIPT roman_Pl end_POSTSUBSCRIPT = 1 such that, making use of Eq. 2.4, the previous equation simplifies to

26Γ2dΓdN+Γ=dlnVdϕ.26superscriptΓ2derivative𝑁ΓΓderivativeitalic-ϕ𝑉\dfrac{2}{6-\Gamma^{2}}\derivative{\Gamma}{N}+\Gamma=-\derivative{\ln V}{\phi}\,.divide start_ARG 2 end_ARG start_ARG 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG divide start_ARG roman_d start_ARG roman_Γ end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG + roman_Γ = - divide start_ARG roman_d start_ARG roman_ln italic_V end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG . (2.7)

As discussed in Ref. [18], this equation is similar to the one of a relativistic particle in presence of friction and accelerated by an external force created by the potential W(ϕ)=ln[V(ϕ)]𝑊italic-ϕ𝑉italic-ϕW(\phi)=\ln[V(\phi)]italic_W ( italic_ϕ ) = roman_ln [ italic_V ( italic_ϕ ) ], the value 66\sqrt{6}square-root start_ARG 6 end_ARG giving the maximal possible speed for the field ϕitalic-ϕ\phiitalic_ϕ. Indeed, positivity of Eq. 1.7 enforces that all field trajectories must satisfy ϵ1<3subscriptitalic-ϵ13\epsilon_{1}<3italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT < 3, i.e., |Γ|<6Γ6|\Gamma|<\sqrt{6}| roman_Γ | < square-root start_ARG 6 end_ARG.

2.2 Slow-roll trajectory

There is no known analytical solution of Eq. 2.7 for an unspecified potential V(ϕ)𝑉italic-ϕV(\phi)italic_V ( italic_ϕ ), although various approximated solutions, in different regimes, have been derived [18] (see, however, Appendix A). In the slow-roll regime we are interested in, one can remark that the field acceleration can also be expressed in terms of ϵ2subscriptitalic-ϵ2\epsilon_{2}italic_ϵ start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT. From Eqs. 2.5 and 1.1, one has

dΓdN=12ϵ2Γ,derivative𝑁Γ12subscriptitalic-ϵ2Γ\derivative{\Gamma}{N}=\dfrac{1}{2}\epsilon_{2}\Gamma,divide start_ARG roman_d start_ARG roman_Γ end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG = divide start_ARG 1 end_ARG start_ARG 2 end_ARG italic_ϵ start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT roman_Γ , (2.8)

in terms of which Eq. 2.7 reads

(1+ϵ262ϵ1)Γ=dlnVdϕ.1subscriptitalic-ϵ262subscriptitalic-ϵ1Γderivativeitalic-ϕ𝑉\left(1+\dfrac{\epsilon_{2}}{6-2\epsilon_{1}}\right)\Gamma=-\derivative{\ln V}% {\phi}\,.( 1 + divide start_ARG italic_ϵ start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT end_ARG start_ARG 6 - 2 italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT end_ARG ) roman_Γ = - divide start_ARG roman_d start_ARG roman_ln italic_V end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG . (2.9)

Assuming a slowly rolling field evolution implies that all the ϵisubscriptitalic-ϵ𝑖\epsilon_{i}italic_ϵ start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT are small and, at leading order, Eq. 2.9 can be approximated by

ΓΓsrdlnVdϕ,similar-to-or-equalsΓsubscriptΓsrderivativeitalic-ϕ𝑉\Gamma\simeq\Gamma_{\mathrm{sr}}\equiv-\derivative{\ln V}{\phi},roman_Γ ≃ roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ≡ - divide start_ARG roman_d start_ARG roman_ln italic_V end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG , (2.10)

which has the solution

Nsr(ϕ)=ϕV(ψ)V(ψ)dψ.subscript𝑁sritalic-ϕsuperscriptitalic-ϕ𝑉𝜓superscript𝑉𝜓𝜓N_{\mathrm{sr}}(\phi)=-\int^{\phi}\dfrac{V(\psi)}{V^{\prime}(\psi)}% \differential{\psi}.italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) = - ∫ start_POSTSUPERSCRIPT italic_ϕ end_POSTSUPERSCRIPT divide start_ARG italic_V ( italic_ψ ) end_ARG start_ARG italic_V start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG . (2.11)

Here the prime stands for the derivative with respect to field value. Another way to interpret this slow-roll trajectory is to remark that the acceleration term of Eq. 2.7 is ignored, which means that we are only considering the friction dominated regime. In fact, were the force term on the right-hand-side be constant, Eq. 2.10 would give the exact terminal velocity of Eq. 2.7, and, Eq. 2.11 would be the exact attractor solution. In the general case, however, there is a small drift sourced by the non-constancy of the force term and this induces differences between Nsr(ϕ)subscript𝑁sritalic-ϕN_{\mathrm{sr}}(\phi)italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) and the exact attractor solution N(ϕ)𝑁italic-ϕN(\phi)italic_N ( italic_ϕ ) of Eq. 2.7.

Let us remark that Eq. 2.11 is defined up to a constant term. However, as explained in Section 1, the quantity of interest for inflation is ΔNΔsubscript𝑁\Delta N_{*}roman_Δ italic_N start_POSTSUBSCRIPT ∗ end_POSTSUBSCRIPT and only the functional ΔN(ϕ)=N(ϕ)NendΔ𝑁italic-ϕ𝑁italic-ϕsubscript𝑁end\Delta N(\phi)=N(\phi)-N_{\mathrm{end}}roman_Δ italic_N ( italic_ϕ ) = italic_N ( italic_ϕ ) - italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT, in which a possible constant term cancels, is observable. As such, in addition to Nsr(ϕ)subscript𝑁sritalic-ϕN_{\mathrm{sr}}(\phi)italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ), one should also estimate Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT accurately.

2.3 Characterizing the end of inflation

By definition, inflation stands for an accelerated expansion of the spacetime, i.e., a¨>0¨𝑎0\ddot{a}>0over¨ start_ARG italic_a end_ARG > 0. From Eq. 2.4, using H=a˙/a𝐻˙𝑎𝑎H=\dot{a}/aitalic_H = over˙ start_ARG italic_a end_ARG / italic_a, one has ϵ1=1a¨/(aH2)subscriptitalic-ϵ11¨𝑎𝑎superscript𝐻2\epsilon_{1}=1-\ddot{a}/(aH^{2})italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT = 1 - over¨ start_ARG italic_a end_ARG / ( italic_a italic_H start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ) and the condition for acceleration translates into ϵ1<1subscriptitalic-ϵ11\epsilon_{1}<1italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT < 1. In the vanilla single-field inflationary models, the accelerated expansion ends by itself with a so-called “graceful exit”: the potential becomes steeper, and the field accelerates up to the point at which

ϵ1endϵ1(Nend)=1.subscriptitalic-ϵ1endsubscriptitalic-ϵ1subscript𝑁end1\epsilon_{1\mathrm{end}}\equiv\epsilon_{1}(N_{\mathrm{end}})=1.italic_ϵ start_POSTSUBSCRIPT 1 roman_e roman_n roman_d end_POSTSUBSCRIPT ≡ italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) = 1 . (2.12)

Translated into velocities, one therefore has ΓendΓ(Nend)=±2subscriptΓendΓsubscript𝑁endplus-or-minus2\Gamma_{\mathrm{end}}\equiv\Gamma(N_{\mathrm{end}})=\pm\sqrt{2}roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ≡ roman_Γ ( italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) = ± square-root start_ARG 2 end_ARG, the sign being related to the direction in which inflation proceeds. Indeed, depending on the shape of the potential, either the field increases during inflation and Γ>0Γ0\Gamma>0roman_Γ > 0, or it decreases and Γ<0Γ0\Gamma<0roman_Γ < 0. It is also possible that inflation ends by another mechanism than a graceful exit, as for instance by a tachyonic instability triggered by an extra field, as in the prototypical hybrid inflation model [91]. In that situation, Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT is no longer set by the condition ϵ1(Nend)=1subscriptitalic-ϵ1subscript𝑁end1\epsilon_{1}(N_{\mathrm{end}})=1italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT ( italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) = 1. Instead, it may be viewed as an additional model parameter. The determination of Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT in these situations has, therefore, nothing to do with the inflationary dynamics, and we will not consider these cases. Let us however stress that a tachyonic instability is relevant only if it affects the inflaton while ϵ1<1subscriptitalic-ϵ11\epsilon_{1}<1italic_ϵ start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT < 1, otherwise it would rather be interpreted as an event belonging to the reheating era.

Solving Eq. 2.12 for Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT is problematic in various aspects. It is a condition on the first Hubble-flow function, or equivalently, on the field velocity ΓΓ\Gammaroman_Γ, whose e-fold dependency would require to solve Eq. 2.7 exactly. Without knowing the exact solution, the best one can do is to use the approximation of Eq. 2.10 and solve Γsr(Nendsr)=±2subscriptΓsrsuperscriptsubscript𝑁endsrplus-or-minus2\Gamma_{\mathrm{sr}}(N_{\mathrm{end}}^{\mathrm{sr}})=\pm\sqrt{2}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ) = ± square-root start_ARG 2 end_ARG instead of Γ(Nend)=±2Γsubscript𝑁endplus-or-minus2\Gamma(N_{\mathrm{end}})=\pm\sqrt{2}roman_Γ ( italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) = ± square-root start_ARG 2 end_ARG. However, by doing so, we break our working hypothesis that the ϵisubscriptitalic-ϵ𝑖\epsilon_{i}italic_ϵ start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT functions have to be small, as the end of inflation is indeed manifestly violating slow roll.

In spite of this, in essentially all works on slow-roll inflation, Nendsrsuperscriptsubscript𝑁endsrN_{\mathrm{end}}^{\mathrm{sr}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT is the value actually used for Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. As we will demonstrate in Section 3, this is quite a good approximation because Nendsrsuperscriptsubscript𝑁endsrN_{\mathrm{end}}^{\mathrm{sr}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT turns out to be the leading order solution of yet another expansion of the field trajectory valid even when slow roll is violated. Another more intuitive explanation justifying the extrapolation of ΓΓsrsimilar-to-or-equalsΓsubscriptΓsr\Gamma\simeq\Gamma_{\mathrm{sr}}roman_Γ ≃ roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT to the end of inflation is to remark that when slow-roll is violated inflation cannot be sustained for a long time, typically not more than (1)order1\order{1}( start_ARG 1 end_ARG ) e-fold. As such, one cannot make a larger error than that on Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT by using Nendsrsuperscriptsubscript𝑁endsrN_{\mathrm{end}}^{\mathrm{sr}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT instead.

In practice, solving Γsr(Nendsr)=±2subscriptΓsrsuperscriptsubscript𝑁endsrplus-or-minus2\Gamma_{\mathrm{sr}}(N_{\mathrm{end}}^{\mathrm{sr}})=\pm\sqrt{2}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ) = ± square-root start_ARG 2 end_ARG consists in finding the root ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT of the algebraic equation

Γsr(ϕendsr)=dlnV(ϕ)dϕ|ϕendsr=±2,subscriptΓsrsuperscriptsubscriptitalic-ϕendsrevaluated-atderivativeitalic-ϕ𝑉italic-ϕsuperscriptsubscriptitalic-ϕendsrplus-or-minus2\Gamma_{\mathrm{sr}}(\phi_{\mathrm{end}}^{\mathrm{sr}})=-\evaluated{% \derivative{\ln V(\phi)}{\phi}}_{\phi_{\mathrm{end}}^{\mathrm{sr}}}=\pm\sqrt{2},roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ) = - start_ARG divide start_ARG roman_d start_ARG roman_ln italic_V ( italic_ϕ ) end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG end_ARG | start_POSTSUBSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_POSTSUBSCRIPT = ± square-root start_ARG 2 end_ARG , (2.13)

and injecting it into the slow-roll trajectory of Eq. 2.11, i.e., Nendsr=Nsr(ϕendsr)superscriptsubscript𝑁endsrsubscript𝑁srsuperscriptsubscriptitalic-ϕendsrN_{\mathrm{end}}^{\mathrm{sr}}=N_{\mathrm{sr}}(\phi_{\mathrm{end}}^{\mathrm{sr% }})italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT = italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ).

The slow-roll approximated trajectory, complemented by its extrapolation to determine the e-fold at which inflation ends, finally gives

ΔNsr(ϕ)Nsr(ϕ)Nendsr=ϕϕendsrV(ψ)V(ψ)dψ,Δsubscript𝑁sritalic-ϕsubscript𝑁sritalic-ϕsuperscriptsubscript𝑁endsrsuperscriptsubscriptitalic-ϕsuperscriptsubscriptitalic-ϕendsr𝑉𝜓superscript𝑉𝜓𝜓\Delta N_{\mathrm{sr}}(\phi)\equiv N_{\mathrm{sr}}(\phi)-N_{\mathrm{end}}^{% \mathrm{sr}}=\int_{\phi}^{\phi_{\mathrm{end}}^{\mathrm{sr}}}\dfrac{V(\psi)}{V^% {\prime}(\psi)}\differential{\psi},roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) ≡ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) - italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT = ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_POSTSUPERSCRIPT divide start_ARG italic_V ( italic_ψ ) end_ARG start_ARG italic_V start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG , (2.14)

where ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT solves Eq. 2.13. The function ΔNsr(ϕ)Δsubscript𝑁sritalic-ϕ\Delta N_{\mathrm{sr}}(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) is the one commonly used to solve the reheating Eq. 1.9. Let us now discuss its accuracy.

2.4 Assessing slow-roll accuracy

In this section, we compare, for various potentials, the slow-roll approximated trajectory ΔNsr(ϕ)Δsubscript𝑁sritalic-ϕ\Delta N_{\mathrm{sr}}(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) defined by Eq. 2.14, to an exact numerical integration of Eq. 2.7 complemented by a root finding algorithm to numerically determine ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT, the solution of Γ(ϕend)=±2Γsubscriptitalic-ϕendplus-or-minus2\Gamma(\phi_{\mathrm{end}})=\pm\sqrt{2}roman_Γ ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) = ± square-root start_ARG 2 end_ARG.

As discussed in the previous sections, the errors made by using ΔNsr(ϕ)Δsubscript𝑁sritalic-ϕ\Delta N_{\mathrm{sr}}(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) instead of ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ) come from both the approximation ΓΓsrsimilar-to-or-equalsΓsubscriptΓsr\Gamma\simeq\Gamma_{\mathrm{sr}}roman_Γ ≃ roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT and ϕendϕendsrsimilar-to-or-equalssubscriptitalic-ϕendsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}\simeq\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ≃ italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT. In order to separate both, let us define a semi-numerical solution, built upon the slow-roll trajectory, based on the exact field value for the end of inflation

ΔNsreeNsr(ϕ)Nsr(ϕend).Δsuperscriptsubscript𝑁sreesubscript𝑁sritalic-ϕsubscript𝑁srsubscriptitalic-ϕend\Delta N_{\mathrm{sr}}^{\mathrm{ee}}\equiv N_{\mathrm{sr}}(\phi)-N_{\mathrm{sr% }}(\phi_{\mathrm{end}}).roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT ≡ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) - italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) . (2.15)

All these functions take as input the field value ϕitalic-ϕ\phiitalic_ϕ and return some approximated number of e-folds. Once we have (numerically) integrated the field trajectory exactly, we have at our disposal the functions N(ϕ)𝑁italic-ϕN(\phi)italic_N ( italic_ϕ ), ϕ(N)italic-ϕ𝑁\phi(N)italic_ϕ ( italic_N ) as well as the value of ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT and Nend=N(ϕend)subscript𝑁end𝑁subscriptitalic-ϕendN_{\mathrm{end}}=N(\phi_{\mathrm{end}})italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = italic_N ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ). From these, we can numerically determine the exact functions ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ) and ϕ(ΔN)italic-ϕΔ𝑁\phi(\Delta N)italic_ϕ ( roman_Δ italic_N ).

Starting from some initial conditions, at ϕ=ϕiniitalic-ϕsubscriptitalic-ϕini\phi=\phi_{\mathrm{ini}}italic_ϕ = italic_ϕ start_POSTSUBSCRIPT roman_ini end_POSTSUBSCRIPT and Nini=0subscript𝑁ini0N_{\mathrm{ini}}=0italic_N start_POSTSUBSCRIPT roman_ini end_POSTSUBSCRIPT = 0, a first approach would be to compare the exact solution N(ϕ)𝑁italic-ϕN(\phi)italic_N ( italic_ϕ ) to its slow-roll approximated version Nsr(ϕ)subscript𝑁sritalic-ϕN_{\mathrm{sr}}(\phi)italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ). Equivalently, one could also compare ϕ(N)italic-ϕ𝑁\phi(N)italic_ϕ ( italic_N ) to ϕsr(N)superscriptitalic-ϕsr𝑁\phi^{\mathrm{sr}}(N)italic_ϕ start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ( italic_N ). Let us first remark that the field value ϕinisubscriptitalic-ϕini\phi_{\mathrm{ini}}italic_ϕ start_POSTSUBSCRIPT roman_ini end_POSTSUBSCRIPT plays no role as, for a given potential, and once on the attractor, the trajectory ϕ(N)italic-ϕ𝑁\phi(N)italic_ϕ ( italic_N ) is universal and always ends in the same manner. Intuitively, one expects Nsr(ϕ)N(ϕ)subscript𝑁sritalic-ϕ𝑁italic-ϕN_{\mathrm{sr}}(\phi)-N(\phi)italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) - italic_N ( italic_ϕ ), as well as ϕsr(N)ϕ(N)superscriptitalic-ϕsr𝑁italic-ϕ𝑁\phi^{\mathrm{sr}}(N)-\phi(N)italic_ϕ start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ( italic_N ) - italic_ϕ ( italic_N ), to be small deep in slow-roll while growing towards the end of inflation and this is exactly what happens. However, we have chosen not to show these trajectories in the following. Indeed, as discussed at length in the introduction, the observable quantity entering the reheating equation is ΔN=N(ϕ)NendΔ𝑁𝑁italic-ϕsubscript𝑁end\Delta N=N(\phi)-N_{\mathrm{end}}roman_Δ italic_N = italic_N ( italic_ϕ ) - italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT, in reference to the end of inflation. As such, any errors damaging the actual value of Nendsubscript𝑁endN_{\mathrm{end}}italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT, as the ones building up close to the end of inflation, will be necessarily folded into all the values of ΔNΔ𝑁\Delta Nroman_Δ italic_N, even if N(ϕ)𝑁italic-ϕN(\phi)italic_N ( italic_ϕ ) and Nsr(ϕ)subscript𝑁sritalic-ϕN_{\mathrm{sr}}(\phi)italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) match well in those regions. Hence, it is actually much more informative to compare ΔNsr(ϕ)Δsubscript𝑁sritalic-ϕ\Delta N_{\mathrm{sr}}(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ), ΔNsree(ϕ)Δsuperscriptsubscript𝑁sreeitalic-ϕ\Delta N_{\mathrm{sr}}^{\mathrm{ee}}(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT ( italic_ϕ ) to the exact ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ), all of these quantities being sensitive to the end of inflation.

Last but not least, the values taken by ϕitalic-ϕ\phiitalic_ϕ are also not very much informative as observable predictions are mostly sensitive to e-fold numbers. Knowing the exact trajectory ϕ(ΔN)italic-ϕΔ𝑁\phi(\Delta N)italic_ϕ ( roman_Δ italic_N ), we can easily trade ϕitalic-ϕ\phiitalic_ϕ for ΔNΔ𝑁\Delta Nroman_Δ italic_N and discuss all error made in terms of the latter quantity. This is relevant because the prototypical value of ΔNN061.5similar-to-or-equalsΔ𝑁subscript𝑁0similar-to-or-equals61.5\Delta N\simeq N_{\scriptscriptstyle{0}}\simeq-61.5roman_Δ italic_N ≃ italic_N start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT ≃ - 61.5 and minimizing errors is particularly important around these figures rather than towards the end of inflation, or, much earlier.

Refer to caption
Figure 1: Absolute error, in e-folds, of the slow-roll approximated trajectory (in red) with respect to the exact value of ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ) for various prototypical models of inflation. The blue curve shows ΔNsree(ϕ)ΔN(ϕ)Δsuperscriptsubscript𝑁sreeitalic-ϕΔ𝑁italic-ϕ\Delta N_{\mathrm{sr}}^{\mathrm{ee}}(\phi)-\Delta N(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT ( italic_ϕ ) - roman_Δ italic_N ( italic_ϕ ) where ΔNsree=Nsr(ϕ)Nsr(ϕend)Δsuperscriptsubscript𝑁sreesubscript𝑁sritalic-ϕsubscript𝑁srsubscriptitalic-ϕend\Delta N_{\mathrm{sr}}^{\mathrm{ee}}=N_{\mathrm{sr}}(\phi)-N_{\mathrm{sr}}(% \phi_{\mathrm{end}})roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT = italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) - italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ), ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT being the exact field value at which inflation stops. The differences between the red and blue curves are the errors induced by using ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT instead of ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT (see text). Let us notice that the Pseudo Natural Inflationary model (lower right) is an extreme case as it has its parameters purposely chosen to be in a slow-roll violating regime (incompatible with current data).

In Fig. 1, we have therefore represented, for various models, as a function of ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ), the absolute errors in the number of e-folds made by using the slow-roll approximated trajectories instead of the exact one. The red curve in these plots shows ΔNsr(ϕ)ΔN(ϕ)Δsubscript𝑁sritalic-ϕΔ𝑁italic-ϕ\Delta N_{\mathrm{sr}}(\phi)-\Delta N(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) - roman_Δ italic_N ( italic_ϕ ) whereas the blue curve is for the semi-analytical trajectory ΔNsree(ϕ)ΔN(ϕ)Δsuperscriptsubscript𝑁sreeitalic-ϕΔ𝑁italic-ϕ\Delta N_{\mathrm{sr}}^{\mathrm{ee}}(\phi)-\Delta N(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT ( italic_ϕ ) - roman_Δ italic_N ( italic_ϕ ). The differences between the blue and red curves are thus coming from the uncertainties in determining the field value at which inflation ends.

The six models considered in Fig. 1 are a few representative of the ones discussed in the Encyclopædia Inflationaris paper of Ref. [14]. We have picked up a quadratic large field inflation model (LFI2subscriptLFI2\mathrm{LFI}_{2}roman_LFI start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT), having a potential V(ϕ)ϕ2proportional-to𝑉italic-ϕsuperscriptitalic-ϕ2V(\phi)\propto\phi^{2}italic_V ( italic_ϕ ) ∝ italic_ϕ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT, a quartic small field inflation model (SFI4lsubscriptSFI4l\mathrm{SFI}_{4\mathrm{l}}roman_SFI start_POSTSUBSCRIPT 4 roman_l end_POSTSUBSCRIPT) with V(ϕ)1(ϕ/μ)4proportional-to𝑉italic-ϕ1superscriptitalic-ϕ𝜇4V(\phi)\propto 1-(\phi/\mu)^{4}italic_V ( italic_ϕ ) ∝ 1 - ( italic_ϕ / italic_μ ) start_POSTSUPERSCRIPT 4 end_POSTSUPERSCRIPT where the vacuum expectation value μ=10𝜇10\mu=10italic_μ = 10 is super-Planckian, Starobinsky Inflation (SISI\mathrm{SI}roman_SI) having V1/2(ϕ)1e2/3ϕproportional-tosuperscript𝑉12italic-ϕ1superscript𝑒23italic-ϕV^{1/2}(\phi)\propto 1-e^{-\sqrt{2/3}\phi}italic_V start_POSTSUPERSCRIPT 1 / 2 end_POSTSUPERSCRIPT ( italic_ϕ ) ∝ 1 - italic_e start_POSTSUPERSCRIPT - square-root start_ARG 2 / 3 end_ARG italic_ϕ end_POSTSUPERSCRIPT, a quadratic T-model inflation (TMITMI\mathrm{TMI}roman_TMI) with V(ϕ)tanh2(ϕ/6)proportional-to𝑉italic-ϕsuperscript2italic-ϕ6V(\phi)\propto\tanh^{2}(\phi/\sqrt{6})italic_V ( italic_ϕ ) ∝ roman_tanh start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ / square-root start_ARG 6 end_ARG ), an exponential supersymmetric inflation model (ESI1subscriptESI1\mathrm{ESI}_{1}roman_ESI start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT) with V(ϕ)1eϕproportional-to𝑉italic-ϕ1superscript𝑒italic-ϕV(\phi)\propto 1-e^{-\phi}italic_V ( italic_ϕ ) ∝ 1 - italic_e start_POSTSUPERSCRIPT - italic_ϕ end_POSTSUPERSCRIPT and a pseudo-natural inflation model (PSNIPSNI\mathrm{PSNI}roman_PSNI) having a potential V(ϕ)1+αln(cosϕ)proportional-to𝑉italic-ϕ1𝛼italic-ϕV(\phi)\propto 1+\alpha\ln(\cos\phi)italic_V ( italic_ϕ ) ∝ 1 + italic_α roman_ln ( start_ARG roman_cos italic_ϕ end_ARG ) with α=1/10𝛼110\alpha=1/10italic_α = 1 / 10. Not all of these models are compatible with current cosmological data, for instance LFI2subscriptLFI2\mathrm{LFI}_{2}roman_LFI start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT is strongly disfavoured whereas PSNIPSNI\mathrm{PSNI}roman_PSNI has its parameters purposely chosen to violate slow-roll (ϵ20.2similar-to-or-equalssubscriptitalic-ϵ20.2\epsilon_{2}\simeq 0.2italic_ϵ start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT ≃ 0.2 during inflation). The small field scenario is a model which is compatible with the data whereas SISI\mathrm{SI}roman_SI, TMITMI\mathrm{TMI}roman_TMI and ESI1subscriptESI1\mathrm{ESI}_{1}roman_ESI start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT are different incarnations of the so-called plateau-type models and belong to most favoured scenarios [19].

Let us remark in Fig. 1 that, for all models, the errors generated by the slow-roll trajectory of Eq. 2.11 grow with the number of e-folds before the end of inflation. This growth is precisely due to the small terms omitted by making the assumption ΓΓsrsimilar-to-or-equalsΓsubscriptΓsr\Gamma\simeq\Gamma_{\mathrm{sr}}roman_Γ ≃ roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT and confirms that ΓsrsubscriptΓsr\Gamma_{\mathrm{sr}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT is slightly off-track the true slow rolling attractor velocity. One can also notice that the error jumps quite fast close to ΔN=0Δ𝑁0\Delta N=0roman_Δ italic_N = 0 whereas, up to one model (PSNIPSNI\mathrm{PSNI}roman_PSNI), it increases like a logarithm at larger values of |ΔN|Δ𝑁|\Delta N|| roman_Δ italic_N |. This is due to the fact that slow-roll is most violated towards the end of inflation and the assumption ΓΓsrsimilar-to-or-equalsΓsubscriptΓsr\Gamma\simeq\Gamma_{\mathrm{sr}}roman_Γ ≃ roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT is quite wrong in these regions. For PSNIPSNI\mathrm{PSNI}roman_PSNI (lower right panel), the errors seem to increase linearly with |ΔN|Δ𝑁|\Delta N|| roman_Δ italic_N |, as opposed to a logarithm-like growth. The reason being that, as aforementioned, it is far from slow roll also during inflation (ϵ2=0.2subscriptitalic-ϵ20.2\epsilon_{2}=0.2italic_ϵ start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT = 0.2).

Finally, these plots confirm that for the fiducial value ΔNN0similar-to-or-equalsΔ𝑁subscript𝑁0\Delta N\simeq N_{\scriptscriptstyle{0}}roman_Δ italic_N ≃ italic_N start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT, the typical errors on the trajectory are (1)order1\order{1}( start_ARG 1 end_ARG ) e-folds. Only for the extreme slow-roll violating model PSNIPSNI\mathrm{PSNI}roman_PSNI, one gets a larger, but still reasonable error.

3 Correcting slow-roll

In this section, we address the main source of error eroding the traditional slow-roll trajectory: ΓsrsubscriptΓsr\Gamma_{\mathrm{sr}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT being slightly off-track the attractor solution.

3.1 Integral constraints

Let us first show that, even though the slow-roll trajectory is not right on the attractor, the shift with respect to the exact solution is actually bounded. One can define the absolute error

ΓΓsr,ΓsubscriptΓsr\mathcal{E}\equiv\Gamma-\Gamma_{\mathrm{sr}},caligraphic_E ≡ roman_Γ - roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT , (3.1)

which can be viewed as a function (ϕ)italic-ϕ\mathcal{E}(\phi)caligraphic_E ( italic_ϕ ) by formally making use of the exact field trajectory for Γ[N(ϕ)]Γdelimited-[]𝑁italic-ϕ\Gamma[N(\phi)]roman_Γ [ italic_N ( italic_ϕ ) ]. From the definition of ΓΓ\Gammaroman_Γ in Eq. 2.5 (still in Planck units) one can rewrite Eq. 2.7 in terms of ϕitalic-ϕ\phiitalic_ϕ as

2Γ6Γ2dΓdϕ+Γ=Γsr.2Γ6superscriptΓ2derivativeitalic-ϕΓΓsubscriptΓsr\dfrac{2\Gamma}{6-\Gamma^{2}}\derivative{\Gamma}{\phi}+\Gamma=\Gamma_{\mathrm{% sr}}.divide start_ARG 2 roman_Γ end_ARG start_ARG 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG divide start_ARG roman_d start_ARG roman_Γ end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG + roman_Γ = roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT . (3.2)

This equation can actually be integrated by separating variables and isolating (ϕ)italic-ϕ\mathcal{E}(\phi)caligraphic_E ( italic_ϕ ) as

ΓΓend2γ6γ2dγ=ϕϕend(ψ)dψ.superscriptsubscriptΓsubscriptΓend2𝛾6superscript𝛾2𝛾superscriptsubscriptitalic-ϕsubscriptitalic-ϕend𝜓𝜓\int_{\Gamma}^{\Gamma_{\mathrm{end}}}\dfrac{2\gamma}{6-\gamma^{2}}% \differential{\gamma}=-\int_{\phi}^{\phi_{\mathrm{end}}}\mathcal{E}(\psi)% \differential{\psi}.∫ start_POSTSUBSCRIPT roman_Γ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG 2 italic_γ end_ARG start_ARG 6 - italic_γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG roman_d start_ARG italic_γ end_ARG = - ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT caligraphic_E ( italic_ψ ) roman_d start_ARG italic_ψ end_ARG . (3.3)

The left-hand side can be integrated exactly and, using Γend2=2superscriptsubscriptΓend22\Gamma_{\mathrm{end}}^{2}=2roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT = 2, one gets the integral constraint

ϕϕend(ψ)dψ=ln[46Γ2(ϕ)].superscriptsubscriptitalic-ϕsubscriptitalic-ϕend𝜓𝜓46superscriptΓ2italic-ϕ\int_{\phi}^{\phi_{\mathrm{end}}}\mathcal{E}(\psi)\differential{\psi}=\ln\left% [\dfrac{4}{6-\Gamma^{2}(\phi)}\right].∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT caligraphic_E ( italic_ψ ) roman_d start_ARG italic_ψ end_ARG = roman_ln [ divide start_ARG 4 end_ARG start_ARG 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG ] . (3.4)

In the slow roll regime Γ21much-less-thansuperscriptΓ21\Gamma^{2}\ll 1roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ≪ 1 and the integrated error made between ϕitalic-ϕ\phiitalic_ϕ and ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT is ln(2/3)0.4similar-to-or-equals230.4\ln(2/3)\simeq-0.4roman_ln ( start_ARG 2 / 3 end_ARG ) ≃ - 0.4. Notice the negative sign, which shows that, for ϕend>ϕsubscriptitalic-ϕenditalic-ϕ\phi_{\mathrm{end}}>\phiitalic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT > italic_ϕ, one has ΓsrΓ>0greater-than-or-equivalent-tosubscriptΓsrΓ0\Gamma_{\mathrm{sr}}\gtrsim\Gamma>0roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ≳ roman_Γ > 0 and the approximated trajectory is slightly advanced compared to the exact one (see also Fig. 1).

We can also derive a second integral constraint by integrating (N)𝑁\mathcal{E}(N)caligraphic_E ( italic_N ) with respect to the number of e-fold. Starting again from Eq. 2.7 and separating the variables ΓΓ\Gammaroman_Γ and N𝑁Nitalic_N, one has

ΓΓend26γ2dγ=NNend(n)dn.superscriptsubscriptΓsubscriptΓend26superscript𝛾2𝛾superscriptsubscript𝑁subscript𝑁end𝑛𝑛\int_{\Gamma}^{\Gamma_{\mathrm{end}}}\dfrac{2}{6-\gamma^{2}}\differential{% \gamma}=-\int_{N}^{N_{\mathrm{end}}}\mathcal{E}(n)\differential{n}.∫ start_POSTSUBSCRIPT roman_Γ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG 2 end_ARG start_ARG 6 - italic_γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG roman_d start_ARG italic_γ end_ARG = - ∫ start_POSTSUBSCRIPT italic_N end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT caligraphic_E ( italic_n ) roman_d start_ARG italic_n end_ARG . (3.5)

Again, the left-hand side can be integrated exactly while the right hand side can be expressed in terms of ϕitalic-ϕ\phiitalic_ϕ by using Eq. 2.5. One obtains another integral constraint

ϕϕend(ψ)Γ(ψ)dψ=16ln[(23)6+Γ(ϕ)6Γ(ϕ)],superscriptsubscriptitalic-ϕsubscriptitalic-ϕend𝜓Γ𝜓𝜓16minus-or-plus236Γitalic-ϕ6Γitalic-ϕ\int_{\phi}^{\phi_{\mathrm{end}}}\dfrac{\mathcal{E}(\psi)}{\Gamma(\psi)}% \differential{\psi}=\dfrac{1}{\sqrt{6}}\ln\left[\left(2\mp\sqrt{3}\right)% \dfrac{\sqrt{6}+\Gamma(\phi)}{\sqrt{6}-\Gamma(\phi)}\right],∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG = divide start_ARG 1 end_ARG start_ARG square-root start_ARG 6 end_ARG end_ARG roman_ln [ ( 2 ∓ square-root start_ARG 3 end_ARG ) divide start_ARG square-root start_ARG 6 end_ARG + roman_Γ ( italic_ϕ ) end_ARG start_ARG square-root start_ARG 6 end_ARG - roman_Γ ( italic_ϕ ) end_ARG ] , (3.6)

where /ΓΓ\mathcal{E}/\Gammacaligraphic_E / roman_Γ is the relative error between the slow-roll and exact trajectory. The ±plus-or-minus\pm± sign is for Γend=±2subscriptΓendplus-or-minus2\Gamma_{\mathrm{end}}=\pm\sqrt{2}roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = ± square-root start_ARG 2 end_ARG, depending on which direction inflation proceeds. Provided |Γ(ϕ)|1much-less-thanΓitalic-ϕ1|\Gamma(\phi)|\ll 1| roman_Γ ( italic_ϕ ) | ≪ 1, the relative integrated error is bounded and reads ln(23)/60.53similar-to-or-equalsminus-or-plus236minus-or-plus0.53\ln(2\mp\sqrt{3})/\sqrt{6}\simeq\mp 0.53roman_ln ( start_ARG 2 ∓ square-root start_ARG 3 end_ARG end_ARG ) / square-root start_ARG 6 end_ARG ≃ ∓ 0.53. We recover that, for ϕend>ϕsubscriptitalic-ϕenditalic-ϕ\phi_{\mathrm{end}}>\phiitalic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT > italic_ϕ, <00\mathcal{E}<0caligraphic_E < 0 and ΓsrΓ>0greater-than-or-equivalent-tosubscriptΓsrΓ0\Gamma_{\mathrm{sr}}\gtrsim\Gamma>0roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ≳ roman_Γ > 0.

Both Eqs. 3.4 and 3.6 are finite and shows that both the absolute error \mathcal{E}caligraphic_E and the relative error /ΓΓ\mathcal{E}/\Gammacaligraphic_E / roman_Γ are under control, deep in the slow roll regime as well as at the end of inflation when ΓΓ\Gammaroman_Γ approaches ΓendsubscriptΓend\Gamma_{\mathrm{end}}roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. This suggests using either (ϕ)italic-ϕ\mathcal{E}(\phi)caligraphic_E ( italic_ϕ ) or (ϕ)/Γ(ϕ)italic-ϕΓitalic-ϕ\mathcal{E}(\phi)/\Gamma(\phi)caligraphic_E ( italic_ϕ ) / roman_Γ ( italic_ϕ ) as a small parameter against which the exact solution of Eq. 2.7 can be expanded.

3.2 New expansion for the field trajectory

From the exact field velocity Γ(ϕ)Γitalic-ϕ\Gamma(\phi)roman_Γ ( italic_ϕ ), the true number of e-fold is given, up to a constant, by

N(ϕ)=ϕ1Γ(ψ)dψ.𝑁italic-ϕsuperscriptitalic-ϕ1Γ𝜓𝜓N(\phi)=\int^{\phi}\dfrac{1}{\Gamma(\psi)}\differential{\psi}.italic_N ( italic_ϕ ) = ∫ start_POSTSUPERSCRIPT italic_ϕ end_POSTSUPERSCRIPT divide start_ARG 1 end_ARG start_ARG roman_Γ ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG . (3.7)

Instead, the traditional slow-roll approximation replaces it with Eq. 2.11. Let us use Eq. 3.7 to express the exact field trajectory as

ΔN(ϕ)=N(ϕ)Nend=ϕϕend1Γsr(ψ)Γsr(ψ)Γ(ψ)dψ,Δ𝑁italic-ϕ𝑁italic-ϕsubscript𝑁endsuperscriptsubscriptitalic-ϕsubscriptitalic-ϕend1subscriptΓsr𝜓subscriptΓsr𝜓Γ𝜓𝜓\Delta N(\phi)=N(\phi)-N_{\mathrm{end}}=-\int_{\phi}^{\phi_{\mathrm{end}}}% \dfrac{1}{\Gamma_{\mathrm{sr}}(\psi)}\dfrac{\Gamma_{\mathrm{sr}}(\psi)}{\Gamma% (\psi)}\differential{\psi},roman_Δ italic_N ( italic_ϕ ) = italic_N ( italic_ϕ ) - italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = - ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG 1 end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ψ ) end_ARG divide start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ψ ) end_ARG start_ARG roman_Γ ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG , (3.8)

where we have artificially introduced the known function Γsr(ϕ)subscriptΓsritalic-ϕ\Gamma_{\mathrm{sr}}(\phi)roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) defined in Eq. 2.10. From the definition of \mathcal{E}caligraphic_E in Eq. 3.1, one has

ΓsrΓ=1Γ,subscriptΓsrΓ1Γ\dfrac{\Gamma_{\mathrm{sr}}}{\Gamma}=1-\dfrac{\mathcal{E}}{\Gamma},divide start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT end_ARG start_ARG roman_Γ end_ARG = 1 - divide start_ARG caligraphic_E end_ARG start_ARG roman_Γ end_ARG , (3.9)

which can be plugged into Eq. 3.8 to get

ΔN(ϕ)=ϕϕend1Γsr(ψ)dψ+ϕϕend(ψ)Γ2(ψ)Γ(ψ)Γsr(ψ)dψ,Δ𝑁italic-ϕsuperscriptsubscriptitalic-ϕsubscriptitalic-ϕend1subscriptΓsr𝜓𝜓superscriptsubscriptitalic-ϕsubscriptitalic-ϕend𝜓superscriptΓ2𝜓Γ𝜓subscriptΓsr𝜓𝜓\Delta N(\phi)=-\int_{\phi}^{\phi_{\mathrm{end}}}\dfrac{1}{\Gamma_{\mathrm{sr}% }(\psi)}\differential{\psi}+\int_{\phi}^{\phi_{\mathrm{end}}}\dfrac{\mathcal{E% }(\psi)}{\Gamma^{2}(\psi)}\dfrac{\Gamma(\psi)}{\Gamma_{\mathrm{sr}}(\psi)}% \differential{\psi},roman_Δ italic_N ( italic_ϕ ) = - ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG 1 end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG + ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG divide start_ARG roman_Γ ( italic_ψ ) end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG , (3.10)

the first term giving back the traditional slow-roll approximation. The second term can be further expanded by remarking that

ΓΓsr=11Γ=1+k=1(Γ)k,ΓsubscriptΓsr11Γ1superscriptsubscript𝑘1superscriptΓ𝑘\dfrac{\Gamma}{\Gamma_{\mathrm{sr}}}=\dfrac{1}{1-\dfrac{\mathcal{E}}{\Gamma}}=% 1+\sum_{k=1}^{\infty}\left(\dfrac{\mathcal{E}}{\Gamma}\right)^{k},divide start_ARG roman_Γ end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT end_ARG = divide start_ARG 1 end_ARG start_ARG 1 - divide start_ARG caligraphic_E end_ARG start_ARG roman_Γ end_ARG end_ARG = 1 + ∑ start_POSTSUBSCRIPT italic_k = 1 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT ∞ end_POSTSUPERSCRIPT ( divide start_ARG caligraphic_E end_ARG start_ARG roman_Γ end_ARG ) start_POSTSUPERSCRIPT italic_k end_POSTSUPERSCRIPT , (3.11)

giving a new and exact expansion for the field trajectory

ΔN(ϕ)=ΔNsree(ϕ)+ϕϕend(ψ)Γ2(ψ)+k=2ϕϕend1Γ(ψ)[(ψ)Γ(ψ)]kdψ.Δ𝑁italic-ϕΔsuperscriptsubscript𝑁sreeitalic-ϕsuperscriptsubscriptitalic-ϕsubscriptitalic-ϕend𝜓superscriptΓ2𝜓superscriptsubscript𝑘2superscriptsubscriptitalic-ϕsubscriptitalic-ϕend1Γ𝜓superscriptdelimited-[]𝜓Γ𝜓𝑘𝜓\Delta N(\phi)=\Delta N_{\mathrm{sr}}^{\mathrm{ee}}(\phi)+\int_{\phi}^{\phi_{% \mathrm{end}}}\dfrac{\mathcal{E}(\psi)}{\Gamma^{2}(\psi)}+\sum_{k=2}^{\infty}% \int_{\phi}^{\phi_{\mathrm{end}}}\dfrac{1}{\Gamma(\psi)}\left[\dfrac{\mathcal{% E}(\psi)}{\Gamma(\psi)}\right]^{k}\differential{\psi}.roman_Δ italic_N ( italic_ϕ ) = roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT ( italic_ϕ ) + ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG + ∑ start_POSTSUBSCRIPT italic_k = 2 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT ∞ end_POSTSUPERSCRIPT ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG 1 end_ARG start_ARG roman_Γ ( italic_ψ ) end_ARG [ divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ ( italic_ψ ) end_ARG ] start_POSTSUPERSCRIPT italic_k end_POSTSUPERSCRIPT roman_d start_ARG italic_ψ end_ARG . (3.12)

Let us notice the first term, which is ΔNsreeΔsuperscriptsubscript𝑁sree\Delta N_{\mathrm{sr}}^{\mathrm{ee}}roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT as the field value ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT here has to be the exact one. Quite importantly, this expansion is not based on the usual slow-roll expansion, one does not need to assume |Γ|1much-less-thanΓ1\absolutevalue{\Gamma}\ll 1| start_ARG roman_Γ end_ARG | ≪ 1. Instead, the “small parameter” is the relative error functional, /ΓΓ\mathcal{E}/\Gammacaligraphic_E / roman_Γ, which is ensured to be under control thanks to the integral constraints derived earlier. The benefit of having expanded the exact trajectory as in Eq. 3.12 is that, as we show in the next section, the second term, which acts as a first correction, is exactly calculable.

Strictly speaking, the expansion of Eq. 3.11 is converging only if the relative error |/Γ|<1Γ1\quantity|\mathcal{E}/\Gamma|<1| start_ARG caligraphic_E / roman_Γ end_ARG | < 1. Although this is ensured for most of the inflationary trajectory, in slow roll, it may exceed unity very close to the end of inflation for |Γsr|>2|Γ|22subscriptΓsr2Γsimilar-to-or-equals22|\Gamma_{\mathrm{sr}}|>2|\Gamma|\simeq 2\sqrt{2}| roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT | > 2 | roman_Γ | ≃ 2 square-root start_ARG 2 end_ARG. However, in order to satisfy the integral constraint of Eq. 3.6, the field domain over which this happens must be small (in Planck units). Similarly, Eq. 3.4 gives a constraint on the absolute error \mathcal{E}caligraphic_E over time, this one cannot not be of order unity for more than a fraction of an e-fold. Although these cases are not of immediate interest when considering ΔNN0similar-to-or-equalsΔ𝑁subscript𝑁0\Delta N\simeq N_{\scriptscriptstyle{0}}roman_Δ italic_N ≃ italic_N start_POSTSUBSCRIPT 0 end_POSTSUBSCRIPT, it is interesting to remark that for |/Γ|>1Γ1\left|\mathcal{E}/\Gamma\right|>1| caligraphic_E / roman_Γ | > 1, one has |/Γsr|<1subscriptΓsr1\left|\mathcal{E}/\Gamma_{\mathrm{sr}}\right|<1| caligraphic_E / roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT | < 1 and another expansion can be performed

ΓsrΓ=11+Γsr=1+k=1(1)k(Γsr)k.subscriptΓsrΓ11subscriptΓsr1superscriptsubscript𝑘1superscript1𝑘superscriptsubscriptΓsr𝑘\dfrac{\Gamma_{\mathrm{sr}}}{\Gamma}=\dfrac{1}{1+\dfrac{\mathcal{E}}{\Gamma_{% \mathrm{sr}}}}=1+\sum_{k=1}^{\infty}\quantity(-1)^{k}\quantity(\dfrac{\mathcal% {E}}{\Gamma_{\mathrm{sr}}})^{k}.divide start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT end_ARG start_ARG roman_Γ end_ARG = divide start_ARG 1 end_ARG start_ARG 1 + divide start_ARG caligraphic_E end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT end_ARG end_ARG = 1 + ∑ start_POSTSUBSCRIPT italic_k = 1 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT ∞ end_POSTSUPERSCRIPT ( start_ARG - 1 end_ARG ) start_POSTSUPERSCRIPT italic_k end_POSTSUPERSCRIPT ( start_ARG divide start_ARG caligraphic_E end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT end_ARG end_ARG ) start_POSTSUPERSCRIPT italic_k end_POSTSUPERSCRIPT . (3.13)

Plugging this expression into Eq. 3.8, one gets another exact expansion

ΔN(ϕ)=ΔNsree(ϕ)+ϕϕend(ψ)Γsr2(ψ)k=2ϕϕend(1)kΓsr(ψ)[(ψ)Γsr(ψ)]kdψ,Δ𝑁italic-ϕΔsuperscriptsubscript𝑁sreeitalic-ϕsuperscriptsubscriptitalic-ϕsubscriptitalic-ϕend𝜓superscriptsubscriptΓsr2𝜓superscriptsubscript𝑘2superscriptsubscriptitalic-ϕsubscriptitalic-ϕendsuperscript1𝑘subscriptΓsr𝜓superscriptdelimited-[]𝜓subscriptΓsr𝜓𝑘𝜓\Delta N(\phi)=\Delta N_{\mathrm{sr}}^{\mathrm{ee}}(\phi)+\int_{\phi}^{\phi_{% \mathrm{end}}}\dfrac{\mathcal{E}(\psi)}{\Gamma_{\mathrm{sr}}^{2}(\psi)}-\sum_{% k=2}^{\infty}\int_{\phi}^{\phi_{\mathrm{end}}}\dfrac{\quantity(-1)^{k}}{\Gamma% _{\mathrm{sr}}(\psi)}\left[\dfrac{\mathcal{E}(\psi)}{\Gamma_{\mathrm{sr}}(\psi% )}\right]^{k}\differential{\psi},roman_Δ italic_N ( italic_ϕ ) = roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT ( italic_ϕ ) + ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG - ∑ start_POSTSUBSCRIPT italic_k = 2 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT ∞ end_POSTSUPERSCRIPT ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG ( start_ARG - 1 end_ARG ) start_POSTSUPERSCRIPT italic_k end_POSTSUPERSCRIPT end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ψ ) end_ARG [ divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ψ ) end_ARG ] start_POSTSUPERSCRIPT italic_k end_POSTSUPERSCRIPT roman_d start_ARG italic_ψ end_ARG , (3.14)

which shows that, up to the field value at which inflation ends, the usual slow-roll approximated trajectory ΔNsreeΔsuperscriptsubscript𝑁sree\Delta N_{\mathrm{sr}}^{\mathrm{ee}}roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT remains the leading order term.

3.3 Velocity correction

Let us now assume that we are in the field domain for which |/Γ|<1Γ1\quantity|\mathcal{E}/\Gamma|<1| start_ARG caligraphic_E / roman_Γ end_ARG | < 1. From Eq. 3.2, one has, exactly

(ϕ)Γ2(ϕ)=2Γ(ψ)[6Γ2(ψ)]dΓdϕ,italic-ϕsuperscriptΓ2italic-ϕ2Γ𝜓delimited-[]6superscriptΓ2𝜓derivativeitalic-ϕΓ\dfrac{\mathcal{E}(\phi)}{\Gamma^{2}(\phi)}=-\dfrac{2}{\Gamma(\psi)\left[6-% \Gamma^{2}(\psi)\right]}\derivative{\Gamma}{\phi}\,,divide start_ARG caligraphic_E ( italic_ϕ ) end_ARG start_ARG roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG = - divide start_ARG 2 end_ARG start_ARG roman_Γ ( italic_ψ ) [ 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) ] end_ARG divide start_ARG roman_d start_ARG roman_Γ end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG , (3.15)

such that

ϕϕend(ψ)Γ2(ψ)dψ=16ln[Γ2(ψ)6Γ2(ψ)]16ln[Γend26Γend2]=16ln[2Γ2(ψ)6Γ2(ψ)],superscriptsubscriptitalic-ϕsubscriptitalic-ϕend𝜓superscriptΓ2𝜓𝜓16superscriptΓ2𝜓6superscriptΓ2𝜓16superscriptsubscriptΓend26superscriptsubscriptΓend2162superscriptΓ2𝜓6superscriptΓ2𝜓\int_{\phi}^{\phi_{\mathrm{end}}}\dfrac{\mathcal{E}(\psi)}{\Gamma^{2}(\psi)}% \differential{\psi}=\dfrac{1}{6}\ln\left[\dfrac{\Gamma^{2}(\psi)}{6-\Gamma^{2}% (\psi)}\right]-\dfrac{1}{6}\ln\left[\dfrac{\Gamma_{\mathrm{end}}^{2}}{6-\Gamma% _{\mathrm{end}}^{2}}\right]=\dfrac{1}{6}\ln\left[\dfrac{2\Gamma^{2}(\psi)}{6-% \Gamma^{2}(\psi)}\right],∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG = divide start_ARG 1 end_ARG start_ARG 6 end_ARG roman_ln [ divide start_ARG roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG start_ARG 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG ] - divide start_ARG 1 end_ARG start_ARG 6 end_ARG roman_ln [ divide start_ARG roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG start_ARG 6 - roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG ] = divide start_ARG 1 end_ARG start_ARG 6 end_ARG roman_ln [ divide start_ARG 2 roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG start_ARG 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ψ ) end_ARG ] , (3.16)

where use has been made of Γend2=2superscriptsubscriptΓend22\Gamma_{\mathrm{end}}^{2}=2roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT = 2. As such, the first correction appearing in the expansion of Eq. 3.12 is a simple velocity correction. Even in slow-roll, for |Γ|1much-less-thanΓ1\absolutevalue{\Gamma}\ll 1| start_ARG roman_Γ end_ARG | ≪ 1, we see that this term matters. As a matter of fact, it grows logarithmically when ΓΓ\Gammaroman_Γ becomes small, and it cancels most of the errors associated with ΔNΔNsreeΔ𝑁Δsuperscriptsubscript𝑁sree\Delta N-\Delta N_{\mathrm{sr}}^{\mathrm{ee}}roman_Δ italic_N - roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT.

We can also rewrite Eq. 3.12 as

ΔN(ϕ)=ΔNsr(ϕ)+16ln[2Γ2(ϕ)6Γ2(ϕ)]+[ΔNsree(ϕ)ΔNsr(ϕ)]+k=2ϕϕend1Γ(ψ)[(ψ)Γ(ψ)]kdψ,Δ𝑁italic-ϕΔsubscript𝑁sritalic-ϕ162superscriptΓ2italic-ϕ6superscriptΓ2italic-ϕdelimited-[]Δsuperscriptsubscript𝑁sreeitalic-ϕΔsubscript𝑁sritalic-ϕsuperscriptsubscript𝑘2superscriptsubscriptitalic-ϕsubscriptitalic-ϕend1Γ𝜓superscriptdelimited-[]𝜓Γ𝜓𝑘𝜓\Delta N(\phi)=\Delta N_{\mathrm{sr}}(\phi)+\dfrac{1}{6}\ln\left[\dfrac{2% \Gamma^{2}(\phi)}{6-\Gamma^{2}(\phi)}\right]+\left[\Delta N_{\mathrm{sr}}^{% \mathrm{ee}}(\phi)-\Delta N_{\mathrm{sr}}(\phi)\right]+\sum_{k=2}^{\infty}\int% _{\phi}^{\phi_{\mathrm{end}}}\dfrac{1}{\Gamma(\psi)}\left[\dfrac{\mathcal{E}(% \psi)}{\Gamma(\psi)}\right]^{k}\differential{\psi},roman_Δ italic_N ( italic_ϕ ) = roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) + divide start_ARG 1 end_ARG start_ARG 6 end_ARG roman_ln [ divide start_ARG 2 roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG start_ARG 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG ] + [ roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT ( italic_ϕ ) - roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) ] + ∑ start_POSTSUBSCRIPT italic_k = 2 end_POSTSUBSCRIPT start_POSTSUPERSCRIPT ∞ end_POSTSUPERSCRIPT ∫ start_POSTSUBSCRIPT italic_ϕ end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG 1 end_ARG start_ARG roman_Γ ( italic_ψ ) end_ARG [ divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ ( italic_ψ ) end_ARG ] start_POSTSUPERSCRIPT italic_k end_POSTSUPERSCRIPT roman_d start_ARG italic_ψ end_ARG , (3.17)

to render explicit the deviations with respect to ΔNsr(ϕ)Δsubscript𝑁sritalic-ϕ\Delta N_{\mathrm{sr}}(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ). The third term appears because ΔNsrΔsubscript𝑁sr\Delta N_{\mathrm{sr}}roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT assumes inflation to end at ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT instead of the exact value ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. We will discuss this issue in Section 4.

An additional issue with Eq. 3.17 is that, in principle, one does not know Γ(ϕ)Γitalic-ϕ\Gamma(\phi)roman_Γ ( italic_ϕ ). However, we are interested in ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ) far from the end of inflation, where slow roll is verified. As such, it is perfectly justified to evaluate Γ(ϕ)Γsr(ϕ)similar-to-or-equalsΓitalic-ϕsubscriptΓsritalic-ϕ\Gamma(\phi)\simeq\Gamma_{\mathrm{sr}}(\phi)roman_Γ ( italic_ϕ ) ≃ roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) within the second term of Eq. 3.17. If a higher precision is needed, it is always possible to account for higher-derivative perturbative corrections by determining ΓΓ\Gammaroman_Γ in terms of ΓsrsubscriptΓsr\Gamma_{\mathrm{sr}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT [92, 19]. In fact, it is interesting to compare our velocity refinement, the second term of Eq. 3.17, to these higher-derivative corrections. As shown in Ref. [93], including the next-to-leading order term to determine Γ(Γsr,Γsr)ΓsubscriptΓsrsuperscriptsubscriptΓsr\Gamma(\Gamma_{\mathrm{sr}},\Gamma_{\mathrm{sr}}^{\prime})roman_Γ ( roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT , roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT ) yields a correction to Nsr(ϕ)subscript𝑁sritalic-ϕN_{\mathrm{sr}}(\phi)italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) in ln[Γsr2(ϕ)]/6superscriptsubscriptΓsr2italic-ϕ6\ln\quantity[\Gamma_{\mathrm{sr}}^{2}(\phi)]/6roman_ln [ start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG ] / 6. This can be compared to first term in the right-hand side of Eq. 3.16, i.e., ln{Γ2(ϕ)/[6Γ(ϕ)2]}/6superscriptΓ2italic-ϕdelimited-[]6Γsuperscriptitalic-ϕ26\ln\left\{\Gamma^{2}(\phi)/[6-\Gamma(\phi)^{2}]\right\}/6roman_ln { roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) / [ 6 - roman_Γ ( italic_ϕ ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ] } / 6. Deep in slow roll, using |Γ||Γsr|1similar-to-or-equalsΓsubscriptΓsrmuch-less-than1\absolutevalue{\Gamma}\simeq\absolutevalue{\Gamma_{\mathrm{sr}}}\ll 1| start_ARG roman_Γ end_ARG | ≃ | start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT end_ARG | ≪ 1, it approximates to ln[Γsr2(ϕ)/6]/6superscriptsubscriptΓsr2italic-ϕ66\ln\quantity[\Gamma_{\mathrm{sr}}^{2}(\phi)/6]/6roman_ln [ start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) / 6 end_ARG ] / 6 and this improves the next-to-leading order higher-derivative correction by a constant shift of ln(6)/60.3similar-to-or-equals660.3-\ln(6)/6\simeq-0.3- roman_ln ( start_ARG 6 end_ARG ) / 6 ≃ - 0.3 e-fold. Let us also stress that, as opposed to perturbative higher-order terms, Eq. 3.16 is exact and this is why we can safely incorporate into the velocity correction some effects coming from the end of inflation (the terms involving ΓendsubscriptΓend\Gamma_{\mathrm{end}}roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT).

In order to check the accuracy of Eq. 3.17, we have plotted in Fig. 2 the absolute error ΔNsrvc(ϕ)ΔN(ϕ)Δsuperscriptsubscript𝑁srvcitalic-ϕΔ𝑁italic-ϕ\Delta N_{\mathrm{sr}}^{\mathrm{vc}}(\phi)-\Delta N(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vc end_POSTSUPERSCRIPT ( italic_ϕ ) - roman_Δ italic_N ( italic_ϕ ) (blue curve) where

ΔNsrvc(ϕ)ΔNsr(ϕ)+16ln[2Γsr2(ϕ)6Γsr2(ϕ)].Δsuperscriptsubscript𝑁srvcitalic-ϕΔsubscript𝑁sritalic-ϕ162superscriptsubscriptΓsr2italic-ϕ6superscriptsubscriptΓsr2italic-ϕ\Delta N_{\mathrm{sr}}^{\mathrm{vc}}(\phi)\equiv\Delta N_{\mathrm{sr}}(\phi)+% \dfrac{1}{6}\ln\left[\dfrac{2\Gamma_{\mathrm{sr}}^{2}(\phi)}{6-\Gamma_{\mathrm% {sr}}^{2}(\phi)}\right].roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vc end_POSTSUPERSCRIPT ( italic_ϕ ) ≡ roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) + divide start_ARG 1 end_ARG start_ARG 6 end_ARG roman_ln [ divide start_ARG 2 roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG start_ARG 6 - roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG ] . (3.18)

Let us stress that we have traded ΓΓ\Gammaroman_Γ for ΓsrsubscriptΓsr\Gamma_{\mathrm{sr}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT in this expression. Compared to the traditional slow-roll trajectory (red), it is evident that the velocity corrections erase the logarithmic error growth with respect to ΔNΔ𝑁\Delta Nroman_Δ italic_N by a factor (10)order10\order{10}( start_ARG 10 end_ARG ). In view of such a success, one may be tempted in trying to evaluate the higher order terms of Eq. 3.17 similarly. However, because they involve powers of /ΓΓ\mathcal{E}/\Gammacaligraphic_E / roman_Γ, using Eq. 3.15 does not allow for an exact integration, even though some parts can still be estimated. We have also tried to calculate exactly these terms starting from Eq. 2.7 and this has allowed us to derive a new exact formula, presented in Appendix A, but which would require the knowledge of V(N)𝑉𝑁V(N)italic_V ( italic_N ), which is not usually the case (see, however, Section 4.2).

Moreover, one can see from Fig. 2 that the remaining error (blue curve) ΔNsrvc(ϕ)ΔN(ϕ)Δsuperscriptsubscript𝑁srvcitalic-ϕΔ𝑁italic-ϕ\Delta N_{\mathrm{sr}}^{\mathrm{vc}}(\phi)-\Delta N(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vc end_POSTSUPERSCRIPT ( italic_ϕ ) - roman_Δ italic_N ( italic_ϕ ) is almost stationary (with respect to ΔNΔ𝑁\Delta Nroman_Δ italic_N) and only driven by the higher order terms of Eq. 3.17. One of them, the third one, encodes the inaccuracies due to the value of ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT. We now turn to this question.

4 Pinpointing the end of inflation

As explained in Section 2.3, the uncertainties associated with ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT come from the trading between ΓΓ\Gammaroman_Γ and ΓsrsubscriptΓsr\Gamma_{\mathrm{sr}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT at the end of inflation, when slow-roll is manifestly violated. A first approach may be to use Eq. 3.9 close to the end of inflation and making use of Eq. 3.15 while replacing ΓΓ\Gammaroman_Γ by ΓsrsubscriptΓsr\Gamma_{\mathrm{sr}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT in the derivative. However, we are now in a regime in which ΓΓ\Gammaroman_Γ is not necessarily close to ΓsrsubscriptΓsr\Gamma_{\mathrm{sr}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT and |/Γ|Γ\quantity|\mathcal{E}/\Gamma|| start_ARG caligraphic_E / roman_Γ end_ARG | could also exceed unity. We have checked that the relative error /ΓΓ\mathcal{E}/\Gammacaligraphic_E / roman_Γ indeed exceeds unity at the end of inflation for two of the tested models: ESI1subscriptESI1\mathrm{ESI}_{1}roman_ESI start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT and PSNIPSNI\mathrm{PSNI}roman_PSNI. A similar approach has been discussed in Refs. [94, 95] and it has been shown to reduce the error in determining the end of inflation down to 5%percent55\%5 % for some specific potentials (LFI2subscriptLFI2\mathrm{LFI}_{2}roman_LFI start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT and TMITMI\mathrm{TMI}roman_TMI-like, which both satisfy |/Γ|<1Γ1\quantity|\mathcal{E}/\Gamma|<1| start_ARG caligraphic_E / roman_Γ end_ARG | < 1). In the following, we will present other methods, performing only slightly better (2%percent22\%2 % uncertainties on ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT) but designed to be robust for all slow-roll models.

4.1 Constrained extrapolations

For a given potential, one has a perfect knowledge of Γsr(ϕ)subscriptΓsritalic-ϕ\Gamma_{\mathrm{sr}}(\phi)roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) and ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT, and we would like to have an accurate determination of Γ(ϕ)Γitalic-ϕ\Gamma(\phi)roman_Γ ( italic_ϕ ) close to ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. As such, we could circumvent the determination of Γ(ϕ)Γitalic-ϕ\Gamma(\phi)roman_Γ ( italic_ϕ ) by trying to approximate instead the integral constraints of the error function (ϕ)italic-ϕ\mathcal{E}(\phi)caligraphic_E ( italic_ϕ ) towards the end of inflation.

For instance, using a trapezoidal approximation for the integrals, one can approximate the first constraint Eq. 3.4 at the end of inflation as

ln[46Γ2(ϕendsr)]=ϕendsrϕend(ψ)dψ12[(ϕend)+(ϕendsr)](ϕendϕendsr).46superscriptΓ2superscriptsubscriptitalic-ϕendsrsuperscriptsubscriptsuperscriptsubscriptitalic-ϕendsrsubscriptitalic-ϕend𝜓𝜓12delimited-[]subscriptitalic-ϕendsuperscriptsubscriptitalic-ϕendsrsubscriptitalic-ϕendsuperscriptsubscriptitalic-ϕendsr\ln\left[\dfrac{4}{6-\Gamma^{2}\quantity(\phi_{\mathrm{end}}^{\mathrm{sr}})}% \right]=\int_{\phi_{\mathrm{end}}^{\mathrm{sr}}}^{\phi_{\mathrm{end}}}\mathcal% {E}(\psi)\differential{\psi}\approx\frac{1}{2}\left[\mathcal{E}(\phi_{\mathrm{% end}})+\mathcal{E}\quantity(\phi_{\mathrm{end}}^{\mathrm{sr}})\right]\quantity% (\phi_{\mathrm{end}}-\phi_{\mathrm{end}}^{\mathrm{sr}}).roman_ln [ divide start_ARG 4 end_ARG start_ARG 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) end_ARG ] = ∫ start_POSTSUBSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT caligraphic_E ( italic_ψ ) roman_d start_ARG italic_ψ end_ARG ≈ divide start_ARG 1 end_ARG start_ARG 2 end_ARG [ caligraphic_E ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) + caligraphic_E ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) ] ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT - italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) . (4.1)

Similarly, one may approximate the second constraint Eq. 3.6, as

16ln[(23)6+Γ(ϕendsr)6Γ(ϕendsr)]=ϕendsrϕend(ψ)Γ(ψ)dψ12[(ϕend)Γ(ϕend)+(ϕendsr)Γ(ϕendsr)](ϕendϕendsr).16minus-or-plus236Γsuperscriptsubscriptitalic-ϕendsr6Γsuperscriptsubscriptitalic-ϕendsrsuperscriptsubscriptsuperscriptsubscriptitalic-ϕendsrsubscriptitalic-ϕend𝜓Γ𝜓𝜓12delimited-[]subscriptitalic-ϕendΓsubscriptitalic-ϕendsuperscriptsubscriptitalic-ϕendsrΓsuperscriptsubscriptitalic-ϕendsrsubscriptitalic-ϕendsuperscriptsubscriptitalic-ϕendsr\dfrac{1}{\sqrt{6}}\ln\left[\left(2\mp\sqrt{3}\right)\dfrac{\sqrt{6}+\Gamma% \quantity(\phi_{\mathrm{end}}^{\mathrm{sr}})}{\sqrt{6}-\Gamma\quantity(\phi_{% \mathrm{end}}^{\mathrm{sr}})}\right]=\int_{\phi_{\mathrm{end}}^{\mathrm{sr}}}^% {\phi_{\mathrm{end}}}\dfrac{\mathcal{E}(\psi)}{\Gamma(\psi)}\differential{\psi% }\approx\frac{1}{2}\left[\frac{\mathcal{E}(\phi_{\mathrm{end}})}{\Gamma(\phi_{% \mathrm{end}})}+\frac{\mathcal{E}\quantity(\phi_{\mathrm{end}}^{\mathrm{sr}})}% {\Gamma\quantity(\phi_{\mathrm{end}}^{\mathrm{sr}})}\right]\quantity(\phi_{% \mathrm{end}}-\phi_{\mathrm{end}}^{\mathrm{sr}}).divide start_ARG 1 end_ARG start_ARG square-root start_ARG 6 end_ARG end_ARG roman_ln [ ( 2 ∓ square-root start_ARG 3 end_ARG ) divide start_ARG square-root start_ARG 6 end_ARG + roman_Γ ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) end_ARG start_ARG square-root start_ARG 6 end_ARG - roman_Γ ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) end_ARG ] = ∫ start_POSTSUBSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT divide start_ARG caligraphic_E ( italic_ψ ) end_ARG start_ARG roman_Γ ( italic_ψ ) end_ARG roman_d start_ARG italic_ψ end_ARG ≈ divide start_ARG 1 end_ARG start_ARG 2 end_ARG [ divide start_ARG caligraphic_E ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) end_ARG start_ARG roman_Γ ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) end_ARG + divide start_ARG caligraphic_E ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) end_ARG start_ARG roman_Γ ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) end_ARG ] ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT - italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) . (4.2)

Under these approximations, Eqs. 4.1 and 4.2 only involve two unknown and independent variables ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT and Γ(ϕendsr)Γsuperscriptsubscriptitalic-ϕendsr\Gamma(\phi_{\mathrm{end}}^{\mathrm{sr}})roman_Γ ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ). Indeed, the error function in both points can be written as

(ϕend)=ΓendΓsr(ϕend), and (ϕendsr)=Γ(ϕendsr)Γend,formulae-sequencesubscriptitalic-ϕendsubscriptΓendsubscriptΓsrsubscriptitalic-ϕend and superscriptsubscriptitalic-ϕendsrΓsuperscriptsubscriptitalic-ϕendsrsubscriptΓend\mathcal{E}(\phi_{\mathrm{end}})=\Gamma_{\mathrm{end}}-\Gamma_{\mathrm{sr}}(% \phi_{\mathrm{end}}),\text{ and }\mathcal{E}\quantity(\phi_{\mathrm{end}}^{% \mathrm{sr}})=\Gamma\quantity(\phi_{\mathrm{end}}^{\mathrm{sr}})-\Gamma_{% \mathrm{end}},caligraphic_E ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) = roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT - roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT ) , and caligraphic_E ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) = roman_Γ ( start_ARG italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_ARG ) - roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT , (4.3)

where, as before, Γend=±2subscriptΓendplus-or-minus2\Gamma_{\mathrm{end}}=\pm\sqrt{2}roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT = ± square-root start_ARG 2 end_ARG. This algebraic system can be solved numerically to obtain an estimate of the field value at the end of inflation ϕendπsuperscriptsubscriptitalic-ϕend𝜋\phi_{\mathrm{end}}^{\pi}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT. Let us stress that, according to the previous discussion, the domain in which these equations are solved requires |Γ(ϕendsr)|<|Γend|Γsuperscriptsubscriptitalic-ϕendsrsubscriptΓend|\Gamma(\phi_{\mathrm{end}}^{\mathrm{sr}})|<|\Gamma_{\mathrm{end}}|| roman_Γ ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ) | < | roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT | and, either ϕendπ<ϕendsrsuperscriptsubscriptitalic-ϕend𝜋superscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\pi}<\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT < italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT or ϕendπ>ϕendsrsuperscriptsubscriptitalic-ϕend𝜋superscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\pi}>\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT > italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT, depending on whether inflation proceeds at decreasing or increasing field values, respectively.

Let us notice that the method could also be accommodated with other functional shapes to model (ϕ)italic-ϕ\mathcal{E}(\phi)caligraphic_E ( italic_ϕ ) in the constraint integrals. We have tested some power-law and exponential extrapolations, but they do not perform better than the simple trapezoidal rule presented here. In the next section, we discuss a similar method anchored on a family of exact field trajectories.

4.2 Matching to Mukhanov inflation

Mukhanov inflation is one of the very few inflationary models for which the exact field trajectory is analytically known, and, the only one which exhibits a graceful exit [96].

Without giving details, the potential is parametrized by two constants α𝛼\alphaitalic_α and β𝛽\betaitalic_β and reads, in Planck units [14]

Vm(ϕ)=M4[1β2(1+2α2ϕ3β)2α2α]exp{3β1α[(1+2α2ϕ3β)2(1α)2α1]}.subscript𝑉mitalic-ϕsuperscript𝑀4delimited-[]1𝛽2superscript12𝛼2italic-ϕ3𝛽2𝛼2𝛼3𝛽1𝛼delimited-[]superscript12𝛼2italic-ϕ3𝛽21𝛼2𝛼1V_{\mathrm{m}}(\phi)=M^{4}\left[1-\dfrac{\beta}{2\left(1+\dfrac{2-\alpha}{2}% \dfrac{\phi}{\sqrt{3\beta}}\right)^{\frac{2\alpha}{2-\alpha}}}\right]\exp\left% \{\dfrac{3\beta}{1-\alpha}\left[\left(1+\dfrac{2-\alpha}{2}\dfrac{\phi}{\sqrt{% 3\beta}}\right)^{\frac{2(1-\alpha)}{2-\alpha}}-1\right]\right\}.italic_V start_POSTSUBSCRIPT roman_m end_POSTSUBSCRIPT ( italic_ϕ ) = italic_M start_POSTSUPERSCRIPT 4 end_POSTSUPERSCRIPT [ 1 - divide start_ARG italic_β end_ARG start_ARG 2 ( 1 + divide start_ARG 2 - italic_α end_ARG start_ARG 2 end_ARG divide start_ARG italic_ϕ end_ARG start_ARG square-root start_ARG 3 italic_β end_ARG end_ARG ) start_POSTSUPERSCRIPT divide start_ARG 2 italic_α end_ARG start_ARG 2 - italic_α end_ARG end_POSTSUPERSCRIPT end_ARG ] roman_exp { divide start_ARG 3 italic_β end_ARG start_ARG 1 - italic_α end_ARG [ ( 1 + divide start_ARG 2 - italic_α end_ARG start_ARG 2 end_ARG divide start_ARG italic_ϕ end_ARG start_ARG square-root start_ARG 3 italic_β end_ARG end_ARG ) start_POSTSUPERSCRIPT divide start_ARG 2 ( 1 - italic_α ) end_ARG start_ARG 2 - italic_α end_ARG end_POSTSUPERSCRIPT - 1 ] } . (4.4)

The expression for Γ(ϕ)Γitalic-ϕ\Gamma(\phi)roman_Γ ( italic_ϕ ) is analytically known and reads

Γm2(ϕ)=3β(1+2α2ϕ3β)2α2α.superscriptsubscriptΓm2italic-ϕ3𝛽superscript12𝛼2italic-ϕ3𝛽2𝛼2𝛼\Gamma_{\mathrm{m}}^{2}(\phi)=\dfrac{3\beta}{\left(1+\dfrac{2-\alpha}{2}\dfrac% {\phi}{\sqrt{3\beta}}\right)^{\frac{2\alpha}{2-\alpha}}}\,.roman_Γ start_POSTSUBSCRIPT roman_m end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) = divide start_ARG 3 italic_β end_ARG start_ARG ( 1 + divide start_ARG 2 - italic_α end_ARG start_ARG 2 end_ARG divide start_ARG italic_ϕ end_ARG start_ARG square-root start_ARG 3 italic_β end_ARG end_ARG ) start_POSTSUPERSCRIPT divide start_ARG 2 italic_α end_ARG start_ARG 2 - italic_α end_ARG end_POSTSUPERSCRIPT end_ARG . (4.5)

The field value at which Mukhanov inflation ends is obtained by solving Γm2(ϕendm)=2superscriptsubscriptΓm2superscriptsubscriptitalic-ϕendm2\Gamma_{\mathrm{m}}^{2}(\phi_{\mathrm{end}}^{\mathrm{m}})=2roman_Γ start_POSTSUBSCRIPT roman_m end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT ) = 2 and reads222Another solution exists for α>2𝛼2\alpha>2italic_α > 2 as the potential develops a maximum located at ϕmax=23β/(α2)subscriptitalic-ϕ23𝛽𝛼2\phi_{\max}=2\sqrt{3\beta}/(\alpha-2)italic_ϕ start_POSTSUBSCRIPT roman_max end_POSTSUBSCRIPT = 2 square-root start_ARG 3 italic_β end_ARG / ( italic_α - 2 ) and there is a symmetry ϕ2ϕmaxϕitalic-ϕ2subscriptitalic-ϕitalic-ϕ\phi\to 2\phi_{\max}-\phiitalic_ϕ → 2 italic_ϕ start_POSTSUBSCRIPT roman_max end_POSTSUBSCRIPT - italic_ϕ in these cases.

ϕendm=23β2α[(3β2)2α2α1].superscriptsubscriptitalic-ϕendm23𝛽2𝛼delimited-[]superscript3𝛽22𝛼2𝛼1\phi_{\mathrm{end}}^{\mathrm{m}}=\dfrac{2\sqrt{3\beta}}{2-\alpha}\left[\left(% \dfrac{3\beta}{2}\right)^{\frac{2-\alpha}{2\alpha}}-1\right].italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT = divide start_ARG 2 square-root start_ARG 3 italic_β end_ARG end_ARG start_ARG 2 - italic_α end_ARG [ ( divide start_ARG 3 italic_β end_ARG start_ARG 2 end_ARG ) start_POSTSUPERSCRIPT divide start_ARG 2 - italic_α end_ARG start_ARG 2 italic_α end_ARG end_POSTSUPERSCRIPT - 1 ] . (4.6)

These functions being exact solution of Eq. 2.7, they automatically satisfy all the integral constraints discussed in Section 3.1.

The present problem is to estimate ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT knowing only Γsr(ϕ)subscriptΓsritalic-ϕ\Gamma_{\mathrm{sr}}(\phi)roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) and one can use Mukhanov inflation as a proxy. For instance, one can determine the value of α𝛼\alphaitalic_α and β𝛽\betaitalic_β such that ΓsrmΓsrsimilar-to-or-equalssuperscriptsubscriptΓsrmsubscriptΓsr\Gamma_{\mathrm{sr}}^{\mathrm{m}}\simeq\Gamma_{\mathrm{sr}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT ≃ roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT. If the slow-roll trajectories of Mukhanov inflation and of the model under scrutiny are close, then should also be their respective exact trajectories. As such, one can use ϕendmsuperscriptsubscriptitalic-ϕendm\phi_{\mathrm{end}}^{\mathrm{m}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT as an approximation of the unknown ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT.

In order to determine the two parameters α𝛼\alphaitalic_α and β𝛽\betaitalic_β, one needs two equations, which cannot be provided by the integral constraints of Section 3.1 as they are already verified. Let us match the first and second derivatives of the potential’s logarithm towards the end of inflation by imposing

Γsrm(ϕendsr)=Γend,dΓsrmdϕ|ϕendsr=dΓsrdϕ|ϕendsr.formulae-sequencesuperscriptsubscriptΓsrmsuperscriptsubscriptitalic-ϕendsrsubscriptΓendevaluated-atderivativeitalic-ϕsuperscriptsubscriptΓsrmsuperscriptsubscriptitalic-ϕendsrevaluated-atderivativeitalic-ϕsubscriptΓsrsuperscriptsubscriptitalic-ϕendsr\Gamma_{\mathrm{sr}}^{\mathrm{m}}(\phi_{\mathrm{end}}^{\mathrm{sr}})=\Gamma_{% \mathrm{end}},\qquad\evaluated{\derivative{\Gamma_{\mathrm{sr}}^{\mathrm{m}}}{% \phi}}_{\phi_{\mathrm{end}}^{\mathrm{sr}}}=\evaluated{\derivative{\Gamma_{% \mathrm{sr}}}{\phi}}_{\phi_{\mathrm{end}}^{\mathrm{sr}}}\,.roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ) = roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT , start_ARG divide start_ARG roman_d start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG end_ARG | start_POSTSUBSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_POSTSUBSCRIPT = start_ARG divide start_ARG roman_d start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG end_ARG | start_POSTSUBSCRIPT italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT end_POSTSUBSCRIPT . (4.7)

Here, ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT refers to the slow-roll approximated value, precisely obtained by solving Γsr(ϕendsr)=ΓendsubscriptΓsrsuperscriptsubscriptitalic-ϕendsrsubscriptΓend\Gamma_{\mathrm{sr}}(\phi_{\mathrm{end}}^{\mathrm{sr}})=\Gamma_{\mathrm{end}}roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ) = roman_Γ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT, while the slow-roll velocity Γsrm(ϕ)superscriptsubscriptΓsrmitalic-ϕ\Gamma_{\mathrm{sr}}^{\mathrm{m}}(\phi)roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT ( italic_ϕ ) of Mukhanov inflation has still to be determined. From Eq. 4.4, one has

Γsrm(ϕ)dlnVmdϕ=β3xα(α+6x)3βxx1+α2(2xαβ),superscriptsubscriptΓsrmitalic-ϕderivativeitalic-ϕsubscript𝑉m𝛽3superscript𝑥𝛼𝛼6𝑥3𝛽𝑥superscript𝑥1𝛼22superscript𝑥𝛼𝛽\Gamma_{\mathrm{sr}}^{\mathrm{m}}(\phi)\equiv-\derivative{\ln V_{\mathrm{m}}}{% \phi}=-\sqrt{\dfrac{\beta}{3}}\dfrac{x^{\alpha}\left(\alpha+6x\right)-3\beta x% }{x^{1+\frac{\alpha}{2}}\left(2x^{\alpha}-\beta\right)}\,,roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT ( italic_ϕ ) ≡ - divide start_ARG roman_d start_ARG roman_ln italic_V start_POSTSUBSCRIPT roman_m end_POSTSUBSCRIPT end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG = - square-root start_ARG divide start_ARG italic_β end_ARG start_ARG 3 end_ARG end_ARG divide start_ARG italic_x start_POSTSUPERSCRIPT italic_α end_POSTSUPERSCRIPT ( italic_α + 6 italic_x ) - 3 italic_β italic_x end_ARG start_ARG italic_x start_POSTSUPERSCRIPT 1 + divide start_ARG italic_α end_ARG start_ARG 2 end_ARG end_POSTSUPERSCRIPT ( 2 italic_x start_POSTSUPERSCRIPT italic_α end_POSTSUPERSCRIPT - italic_β ) end_ARG , (4.8)

where

x(ϕ)(1+2α2ϕ3β)22α.𝑥italic-ϕsuperscript12𝛼2italic-ϕ3𝛽22𝛼x(\phi)\equiv\left(1+\dfrac{2-\alpha}{2}\dfrac{\phi}{\sqrt{3\beta}}\right)^{% \frac{2}{2-\alpha}}.italic_x ( italic_ϕ ) ≡ ( 1 + divide start_ARG 2 - italic_α end_ARG start_ARG 2 end_ARG divide start_ARG italic_ϕ end_ARG start_ARG square-root start_ARG 3 italic_β end_ARG end_ARG ) start_POSTSUPERSCRIPT divide start_ARG 2 end_ARG start_ARG 2 - italic_α end_ARG end_POSTSUPERSCRIPT . (4.9)

Taking the derivative of Eq. 4.8 with respect to ϕitalic-ϕ\phiitalic_ϕ gives

dΓsrmdϕ=2(α+2)(α+6x)x2αβ[α(2α)+12x(α+1)]xα+3αβ2x6x2(2xαβ)2.derivativeitalic-ϕsuperscriptsubscriptΓsrm2𝛼2𝛼6𝑥superscript𝑥2𝛼𝛽delimited-[]𝛼2𝛼12𝑥𝛼1superscript𝑥𝛼3𝛼superscript𝛽2𝑥6superscript𝑥2superscript2superscript𝑥𝛼𝛽2\derivative{\Gamma_{\mathrm{sr}}^{\mathrm{m}}}{\phi}=\dfrac{2\left(\alpha+2% \right)\left(\alpha+6x\right)x^{2\alpha}-\beta\left[\alpha\left(2-\alpha\right% )+12x\left(\alpha+1\right)\right]x^{\alpha}+3\alpha\beta^{2}x}{6x^{2}\left(2x^% {\alpha}-\beta\right)^{2}}\,.divide start_ARG roman_d start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG = divide start_ARG 2 ( italic_α + 2 ) ( italic_α + 6 italic_x ) italic_x start_POSTSUPERSCRIPT 2 italic_α end_POSTSUPERSCRIPT - italic_β [ italic_α ( 2 - italic_α ) + 12 italic_x ( italic_α + 1 ) ] italic_x start_POSTSUPERSCRIPT italic_α end_POSTSUPERSCRIPT + 3 italic_α italic_β start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT italic_x end_ARG start_ARG 6 italic_x start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( 2 italic_x start_POSTSUPERSCRIPT italic_α end_POSTSUPERSCRIPT - italic_β ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG . (4.10)

Let us mention that the slow-roll approximated functions of Eqs. 4.8 and 4.10 are also related to the so-called potential slow-roll parameters ϵ1V=(Γsrm)2/2subscriptitalic-ϵ1𝑉superscriptsuperscriptsubscriptΓsrm22\epsilon_{1V}=\left(\Gamma_{\mathrm{sr}}^{\mathrm{m}}\right)^{2}/2italic_ϵ start_POSTSUBSCRIPT 1 italic_V end_POSTSUBSCRIPT = ( roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT ) start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT / 2 and ϵ2V=2dΓsrm/dϕsubscriptitalic-ϵ2𝑉2superscriptsubscriptΓsrmitalic-ϕ\epsilon_{2V}=2\differential{\Gamma_{\mathrm{sr}}^{\mathrm{m}}}/\differential{\phi}italic_ϵ start_POSTSUBSCRIPT 2 italic_V end_POSTSUBSCRIPT = 2 roman_d start_ARG roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT end_ARG / roman_d start_ARG italic_ϕ end_ARG.

For a given inflationary potential V(ϕ)𝑉italic-ϕV(\phi)italic_V ( italic_ϕ ), plugging Eqs. 4.8 and 4.10 into Eq. 4.7 gives a set of two algebraic equations for α𝛼\alphaitalic_α and β𝛽\betaitalic_β that has to be solved numerically. Once the value of α𝛼\alphaitalic_α and β𝛽\betaitalic_β are determined, the exact field value at which Mukhanov inflation ends is given by Eq. 4.6 and this will be taken as the estimator of the unknown ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT.

4.3 End point correction

LFI2subscriptLFI2\mathrm{LFI}_{2}roman_LFI start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT SFI4lsubscriptSFI4l\mathrm{SFI}_{4\mathrm{l}}roman_SFI start_POSTSUBSCRIPT 4 roman_l end_POSTSUBSCRIPT SISI\mathrm{SI}roman_SI TMITMI\mathrm{TMI}roman_TMI ESI1subscriptESI1\mathrm{ESI}_{1}roman_ESI start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT PSNIPSNI\mathrm{PSNI}roman_PSNI
ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT 1.0091.0091.0091.009 0%percent00\%0 % 9.6579.6579.6579.657 0%percent00\%0 % 0.6150.6150.6150.615 0%percent00\%0 % 0.8390.8390.8390.839 0%percent00\%0 % 0.2710.2710.2710.271 0%percent00\%0 % 1.5641.5641.5641.564 0%percent00\%0 %
ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT 1.4141.4141.4141.414 40%percent4040\%40 % 9.3619.3619.3619.361 3.1%percent3.13.1\%3.1 % 0.9400.9400.9400.940 53%percent5353\%53 % 1.2081.2081.2081.208 44%percent4444\%44 % 0.5350.5350.5350.535 97%percent9797\%97 % 1.4781.4781.4781.478 5%percent55\%5 %
ϕendπsuperscriptsubscriptitalic-ϕend𝜋\phi_{\mathrm{end}}^{\pi}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT 0.9840.9840.9840.984 2.5%percent2.52.5\%2.5 % 9.6619.6619.6619.661 0.04%percent0.040.04\%0.04 % 0.6070.6070.6070.607 1.3%percent1.31.3\%1.3 % 0.8260.8260.8260.826 1.5%percent1.51.5\%1.5 % 0.2730.2730.2730.273 0.7%percent0.70.7\%0.7 % -- --
ϕendmsuperscriptsubscriptitalic-ϕendm\phi_{\mathrm{end}}^{\mathrm{m}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT 0.9860.9860.9860.986 2.3%percent2.32.3\%2.3 % 9.6789.6789.6789.678 0.2%percent0.20.2\%0.2 % 0.5940.5940.5940.594 3.4%percent3.43.4\%3.4 % 0.8250.8250.8250.825 1.7%percent1.71.7\%1.7 % 0.2380.2380.2380.238 12%percent1212\%12 % 1.6041.6041.6041.604 2%percent22\%2 %
Table 1: Comparison of various approximations to determine the field value at which inflation ends. The exact value is ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT, the slow-roll approximation gives ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT, the trapezoid error function extrapolation yields ϕendπsuperscriptsubscriptitalic-ϕend𝜋\phi_{\mathrm{end}}^{\pi}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT, and, matching Mukhanov inflation at ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT gives the value ϕendmsuperscriptsubscriptitalic-ϕendm\phi_{\mathrm{end}}^{\mathrm{m}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT. The number quoted in percent is the relative error in reference to the exact value ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. The gain in precision on ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT by the methods presented here is about an order of magnitude compared to slow-roll.

In Table 1, we give the exact numerical value of ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT, the slow-roll approximated value ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT, the trapezoid-approximated value ϕendπsuperscriptsubscriptitalic-ϕend𝜋\phi_{\mathrm{end}}^{\pi}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT and the Mukhanov-approximated ϕendmsuperscriptsubscriptitalic-ϕendm\phi_{\mathrm{end}}^{\mathrm{m}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT, for the six inflationary models presented in Section 2.4. For the quite extreme case PSNIPSNI\mathrm{PSNI}roman_PSNI, we have not reported the value of ϕendπsuperscriptsubscriptitalic-ϕend𝜋\phi_{\mathrm{end}}^{\pi}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT as we have found more than one numerical solution for Eqs. 4.1 and 4.2 thereby preventing an easy determination of the best estimator. Let us notice that the value of ϕendmsuperscriptsubscriptitalic-ϕendm\phi_{\mathrm{end}}^{\mathrm{m}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT, even if close to the exact one, lies in a domain for which the PSNIPSNI\mathrm{PSNI}roman_PSNI potential is not defined (ϕ>π/2italic-ϕ𝜋2\phi>\pi/2italic_ϕ > italic_π / 2) and, as such, it is certainly not really useful. Let us stress, again, that numerically solving algebraic equations, such as Eqs. 4.1 and 4.2, or, Eq. 4.7, is orders of magnitude faster than numerically integrating Eq. 2.7 all along inflation. As can be seen in this table, the relative error in reference to ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT is reduced by an order of magnitude using either the trapezoidal approximation or the Mukhanov inflation matching method instead of the traditional slow-roll value.

Refer to caption
Figure 2: Absolute error, in e-folds, of the velocity-corrected trajectory ΔNsrvcΔNΔsuperscriptsubscript𝑁srvcΔ𝑁\Delta N_{\mathrm{sr}}^{\mathrm{vc}}-\Delta Nroman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vc end_POSTSUPERSCRIPT - roman_Δ italic_N (blue curve), of the velocity plus end-point corrected trajectories ΔNsrvcmΔNΔsuperscriptsubscript𝑁srvcmΔ𝑁\Delta N_{\mathrm{sr}}^{\mathrm{vcm}}-\Delta Nroman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vcm end_POSTSUPERSCRIPT - roman_Δ italic_N (green curve) and ΔNsrvcπΔNΔsuperscriptsubscript𝑁srvc𝜋Δ𝑁\Delta N_{\mathrm{sr}}^{\mathrm{vc}\pi}-\Delta Nroman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vc italic_π end_POSTSUPERSCRIPT - roman_Δ italic_N (magenta curve), with respect to the exact value of ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ) for various prototypical models of inflation. The red curve is the error associated with the traditional slow-roll approximation, same as in Fig. 1.

One can now define an additional correction to slow roll, including both the velocity correction of Section 3.3 and a better estimation of the end point field value, as

ΔNsrvcπΔsuperscriptsubscript𝑁srvc𝜋\displaystyle\Delta N_{\mathrm{sr}}^{\mathrm{vc}\pi}roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vc italic_π end_POSTSUPERSCRIPT Nsr(ϕ)Nsr(ϕendπ)+16ln[2Γsr2(ϕ)6Γsr2(ϕ)]absentsubscript𝑁sritalic-ϕsubscript𝑁srsuperscriptsubscriptitalic-ϕend𝜋162superscriptsubscriptΓsr2italic-ϕ6superscriptsubscriptΓsr2italic-ϕ\displaystyle\equiv N_{\mathrm{sr}}(\phi)-N_{\mathrm{sr}}(\phi_{\mathrm{end}}^% {\pi})+\dfrac{1}{6}\ln\left[\dfrac{2\Gamma_{\mathrm{sr}}^{2}(\phi)}{6-\Gamma_{% \mathrm{sr}}^{2}(\phi)}\right]≡ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) - italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT ) + divide start_ARG 1 end_ARG start_ARG 6 end_ARG roman_ln [ divide start_ARG 2 roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG start_ARG 6 - roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG ] (4.11)
=ΔNsr(ϕ)+16ln[2Γsr2(ϕ)6Γsr2(ϕ)]+Nsr(ϕendsr)Nsr(ϕendπ),absentΔsubscript𝑁sritalic-ϕ162superscriptsubscriptΓsr2italic-ϕ6superscriptsubscriptΓsr2italic-ϕsubscript𝑁srsuperscriptsubscriptitalic-ϕendsrsubscript𝑁srsuperscriptsubscriptitalic-ϕend𝜋\displaystyle=\Delta N_{\mathrm{sr}}(\phi)+\dfrac{1}{6}\ln\left[\dfrac{2\Gamma% _{\mathrm{sr}}^{2}(\phi)}{6-\Gamma_{\mathrm{sr}}^{2}(\phi)}\right]+N_{\mathrm{% sr}}(\phi_{\mathrm{end}}^{\mathrm{sr}})-N_{\mathrm{sr}}(\phi_{\mathrm{end}}^{% \pi}),= roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) + divide start_ARG 1 end_ARG start_ARG 6 end_ARG roman_ln [ divide start_ARG 2 roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG start_ARG 6 - roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG ] + italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ) - italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_π end_POSTSUPERSCRIPT ) ,

and a similar expression for the Mukhanov-approximated value

ΔNsrvcmΔsuperscriptsubscript𝑁srvcm\displaystyle\Delta N_{\mathrm{sr}}^{\mathrm{vcm}}roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vcm end_POSTSUPERSCRIPT =ΔNsr(ϕ)+16ln[2Γsr2(ϕ)6Γsr2(ϕ)]+Nsr(ϕendsr)Nsr(ϕendm).absentΔsubscript𝑁sritalic-ϕ162superscriptsubscriptΓsr2italic-ϕ6superscriptsubscriptΓsr2italic-ϕsubscript𝑁srsuperscriptsubscriptitalic-ϕendsrsubscript𝑁srsuperscriptsubscriptitalic-ϕendm\displaystyle=\Delta N_{\mathrm{sr}}(\phi)+\dfrac{1}{6}\ln\left[\dfrac{2\Gamma% _{\mathrm{sr}}^{2}(\phi)}{6-\Gamma_{\mathrm{sr}}^{2}(\phi)}\right]+N_{\mathrm{% sr}}(\phi_{\mathrm{end}}^{\mathrm{sr}})-N_{\mathrm{sr}}(\phi_{\mathrm{end}}^{% \mathrm{m}}).= roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ ) + divide start_ARG 1 end_ARG start_ARG 6 end_ARG roman_ln [ divide start_ARG 2 roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG start_ARG 6 - roman_Γ start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_ϕ ) end_ARG ] + italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT ) - italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT ( italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_m end_POSTSUPERSCRIPT ) . (4.12)

The performance of ΔNsrvcπ(ϕ)Δsuperscriptsubscript𝑁srvc𝜋italic-ϕ\Delta N_{\mathrm{sr}}^{\mathrm{vc}\pi}(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vc italic_π end_POSTSUPERSCRIPT ( italic_ϕ ) and ΔNsrvcm(ϕ)Δsuperscriptsubscript𝑁srvcmitalic-ϕ\Delta N_{\mathrm{sr}}^{\mathrm{vcm}}(\phi)roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vcm end_POSTSUPERSCRIPT ( italic_ϕ ) have been represented in Fig. 2, as green and magenta curves, compared to the traditional slow roll ΔNsrΔsubscript𝑁sr\Delta N_{\mathrm{sr}}roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT (red) and to the velocity corrected trajectory ΔNsrvcΔsuperscriptsubscript𝑁srvc\Delta N_{\mathrm{sr}}^{\mathrm{vc}}roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_vc end_POSTSUPERSCRIPT (blue). There is some improvement for LFI2subscriptLFI2\mathrm{LFI}_{2}roman_LFI start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT while for the other model, using a more accurate field value for the end of inflation produces a slight overshoot for the total correction. As can be checked in Eq. 3.17, the next term (k=2𝑘2k=2italic_k = 2) in the expansion has an opposite sign to the third one, ΔNsreeΔNsrΔsuperscriptsubscript𝑁sreeΔsubscript𝑁sr\Delta N_{\mathrm{sr}}^{\mathrm{ee}}-\Delta N_{\mathrm{sr}}roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_ee end_POSTSUPERSCRIPT - roman_Δ italic_N start_POSTSUBSCRIPT roman_sr end_POSTSUBSCRIPT, which is the error induced by using ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT instead ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. As such, they somehow compensate and, for some models, the best option is to keep the simple velocity correction with ϕendsrsuperscriptsubscriptitalic-ϕendsr\phi_{\mathrm{end}}^{\mathrm{sr}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT start_POSTSUPERSCRIPT roman_sr end_POSTSUPERSCRIPT in the trajectory. Notice, however, that a more accurate value for ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT is always beneficial for estimating ρendsubscript𝜌end\rho_{\mathrm{end}}italic_ρ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT while the small overshooting is never large enough to spoil the velocity correction.

5 Conclusion

In this work, we have proposed new methods to improve the analytical observable predictions of the slow-roll single field models of inflation. Complementing most of the works in the literature that have been focused on the generation of cosmological perturbations during inflation, we have been focused here on a quite neglected aspect which concerns the accuracy at which the background field trajectory can be determined. As explained in the introduction, determining with precision the relation ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ) is crucial to correctly map wavenumbers today to wavenumbers during inflation. Moreover, because the reheating era lies in between the standard hot Big-Bang model eras and Cosmic Inflation, any uncertainties on ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ) will bias any inference made onto the kinematics of the reheating era.

One of the main results of this work is the exact expansion of Eq. 3.12, which has allowed us to present a simple and practical velocity correction to the usual slow-roll trajectory, as defined in Eq. 3.18. Adding this correction is trivial and, as shown in Fig. 2, immediately kills the absolute error on ΔN(ϕ)Δ𝑁italic-ϕ\Delta N(\phi)roman_Δ italic_N ( italic_ϕ ) by an order of magnitude, for all the tested models. We have also discussed additional improvements to better determine the field value at which inflation ends. This end-point correction does not necessarily perform better than the velocity correction alone as it breaks some fortuitous compensation of some neglected higher order terms. Nonetheless, it never degrades significantly the velocity-corrected trajectory and always allows for a more accurate determination of ρendsubscript𝜌end\rho_{\mathrm{end}}italic_ρ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT.

Various other new results have been obtained along the course of searching for slow-roll improvements, such as the derivation of new integral constraints in Section 3.1, and, an exact, but still formal, new solution for the field trajectory when the functional V(N)𝑉𝑁V(N)italic_V ( italic_N ) is known, see Appendix A. Our work could be improved in various directions, such as estimating the next terms in the expansion of Eq. 3.18, or, devising more involved methods to determine ϕendsubscriptitalic-ϕend\phi_{\mathrm{end}}italic_ϕ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. However, one should keep in mind that more involved analytical methods are relevant only if they remain simpler, and numerically much faster, than bruteforcely integrating Eq. 2.7. The present work may be precisely filling this niche.

Acknowledgements

We would like thank J. Martin and V. Vennin for enlightening discussions and for providing useful comments on the manuscript. This work is supported by the ESA Belgian Federal PRODEX Grant N4000143201superscriptN4000143201\mathrm{N^{\circ}}4000143201roman_N start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT 4000143201, the Wallonia-Brussels Federation Grant ARC N19/24103superscriptN1924103\mathrm{N^{\circ}}19/24-103roman_N start_POSTSUPERSCRIPT ∘ end_POSTSUPERSCRIPT 19 / 24 - 103. B. B. is publishing in the quality of ASPIRANT Research Fellow of the FNRS.

Appendix A Formal solution

Starting from Eq. 2.7, instead of trying to use the field value ϕitalic-ϕ\phiitalic_ϕ as a variable, one may switch to the number of e-folds N𝑁Nitalic_N and express the right-hand side as

dlnVdϕ=1Γ(N)dlnVdN,derivativeitalic-ϕ𝑉1Γ𝑁derivative𝑁𝑉\derivative{\ln V}{\phi}=\dfrac{1}{\Gamma(N)}\derivative{\ln V}{N}\,,divide start_ARG roman_d start_ARG roman_ln italic_V end_ARG end_ARG start_ARG roman_d start_ARG italic_ϕ end_ARG end_ARG = divide start_ARG 1 end_ARG start_ARG roman_Γ ( italic_N ) end_ARG divide start_ARG roman_d start_ARG roman_ln italic_V end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG , (A.1)

where the potential V(ϕ)𝑉italic-ϕV(\phi)italic_V ( italic_ϕ ) is now viewed as a V(N)=V[ϕ(N)]𝑉𝑁𝑉delimited-[]italic-ϕ𝑁V(N)=V[\phi(N)]italic_V ( italic_N ) = italic_V [ italic_ϕ ( italic_N ) ]. Plugging Eq. A.1 into Eq. 2.7, multiplying both sides by Γ(6Γ2)Γ6superscriptΓ2\Gamma(6-\Gamma^{2})roman_Γ ( 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ) one gets a differential equation for Γ2superscriptΓ2\Gamma^{2}roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT

dΓ2dN+(6Γ2)(Γ2+dlnVdN)=0.derivative𝑁superscriptΓ26superscriptΓ2superscriptΓ2derivative𝑁𝑉0\derivative{\Gamma^{2}}{N}+\left(6-\Gamma^{2}\right)\left(\Gamma^{2}+% \derivative{\ln V}{N}\right)=0.divide start_ARG roman_d start_ARG roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG + ( 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ) ( roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT + divide start_ARG roman_d start_ARG roman_ln italic_V end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG ) = 0 . (A.2)

This differential equation is a non-homogeneous Riccati equation and can be solved analytically [97]. Let us define the new “boost” function

Λ(N)16Γ2(N).Λ𝑁16superscriptΓ2𝑁\Lambda(N)\equiv\dfrac{1}{6-\Gamma^{2}(N)}\,.roman_Λ ( italic_N ) ≡ divide start_ARG 1 end_ARG start_ARG 6 - roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ( italic_N ) end_ARG . (A.3)

In terms of Λ(N)Λ𝑁\Lambda(N)roman_Λ ( italic_N ), Eq. A.2 considerably simplifies into

dΛdN+[6+V(N)V(N)]Λ=1,derivative𝑁Λdelimited-[]6superscript𝑉𝑁𝑉𝑁Λ1\derivative{\Lambda}{N}+\left[6+\dfrac{V^{\prime}(N)}{V(N)}\right]\Lambda=1,divide start_ARG roman_d start_ARG roman_Λ end_ARG end_ARG start_ARG roman_d start_ARG italic_N end_ARG end_ARG + [ 6 + divide start_ARG italic_V start_POSTSUPERSCRIPT ′ end_POSTSUPERSCRIPT ( italic_N ) end_ARG start_ARG italic_V ( italic_N ) end_ARG ] roman_Λ = 1 , (A.4)

This is a non-homogeneous linear differential equation which admits the exact solution

Λ(N)=e6ΔNVendV(N)ΛendNNende6(nN)V(n)V(N)dn,Λ𝑁superscript𝑒6Δ𝑁subscript𝑉end𝑉𝑁subscriptΛendsuperscriptsubscript𝑁subscript𝑁endsuperscript𝑒6𝑛𝑁𝑉𝑛𝑉𝑁𝑛\Lambda(N)=e^{-6\Delta N}\dfrac{V_{\mathrm{end}}}{V(N)}\Lambda_{\mathrm{end}}-% \int_{N}^{N_{\mathrm{end}}}e^{6\left(n-N\right)}\dfrac{V(n)}{V(N)}% \differential{n},roman_Λ ( italic_N ) = italic_e start_POSTSUPERSCRIPT - 6 roman_Δ italic_N end_POSTSUPERSCRIPT divide start_ARG italic_V start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_ARG start_ARG italic_V ( italic_N ) end_ARG roman_Λ start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT - ∫ start_POSTSUBSCRIPT italic_N end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT end_POSTSUPERSCRIPT italic_e start_POSTSUPERSCRIPT 6 ( italic_n - italic_N ) end_POSTSUPERSCRIPT divide start_ARG italic_V ( italic_n ) end_ARG start_ARG italic_V ( italic_N ) end_ARG roman_d start_ARG italic_n end_ARG , (A.5)

where, as before, ΔNNNendΔ𝑁𝑁subscript𝑁end\Delta N\equiv N-N_{\mathrm{end}}roman_Δ italic_N ≡ italic_N - italic_N start_POSTSUBSCRIPT roman_end end_POSTSUBSCRIPT. This solution extends an approximated one derived in Ref. [18] under the “non-relativistic” assumption (Γ26much-less-thansuperscriptΓ26\Gamma^{2}\ll 6roman_Γ start_POSTSUPERSCRIPT 2 end_POSTSUPERSCRIPT ≪ 6).

References