Stochastic tunneling in de Sitter spacetime

Tunneling processes in de Sitter spacetime are studied by using the stochastic approach. We exploit the Martin-Siggia-Rose-Janssen-de Dominicis (MSRJD) functional integral to obtain the tunneling rate. The applicability conditions of this method are clarified using the Schwinger-Keldysh formalism. In the case of a shallow potential barrier, we reproduce the Hawking-Moss (HM) tunneling rate. Remarkably, in contrast to HM picture, the configuration derived from the MSRJD functional integral satisfies physically natural boundary conditions. We also discuss the case of a steep potential barrier and find an interesting Coleman-de Luccia (CDL) bubble-like configuration. Our results demonstrate how the bubble nucleation process could be described in the stochastic approach. Our method turns out to be useful for investigating various tunneling processes during inflation.


Introduction
Tunneling is an important non-perturbative phenomenon in cosmology.Hence, it is crucial to understand tunneling processes in curved spacetime.In particular, tunneling in de Sitter spacetime is worth studying in detail.The reason is that de Sitter spacetime is the simplest non-trivial curved spacetime.Moreover, the results obtained there can be applicable to tunneling phenomena during inflation.Since both Minkowski and de Sitter spacetime have the maximal symmetry, there exist a similarity.Indeed, the Euclidean instanton method for evaluating the tunneling rate in Minkowski spacetime [1,2] is applicable to tunneling processes in de Sitter spacetime [3].However, the extension of Euclidean method to de Sitter spacetime has a difficulty in interpretation.In the case of de Sitter spacetime, there are two saddle solutions: Coleman-De Luccia (CDL) [3] and Hawking-Moss (HM) instantons [4].The latter instanton is a homogeneous solution.Therefore, it is not straightforward to interpret the tunneling rate as that for the tunneling process from a false vacuum to a true vacuum [5,6].
To circumvent the difficulty, a real time formalism might be useful.In the case of de Sitter spacetime in the flat chart, the stochastic formalism has been used for investigating time evolution of fluctuations during inflation [7,8].In this formalism, Langevin equations describe the dynamics of long wavelength modes.Stochastic noises stem from short wavelength quantum modes.The stochastic formalism has been also used to analyze the tunneling processes in de Sitter spacetime [9][10][11][12][13][14][15].However, most of the previous works treated only the HM transition by utilizing the Fokker-Plank (FP) equation for a homogeneous scalar field, which can be derived from the Langevin equation.Obviously, the HM transition cannot describe the bubble nucleation in contrast to CDL tunneling.To describe the bubble, we have to include the spatial gradient term and the FP equation becomes the functional differential equation instead of the partial differential equation.Hence, at first glance, it seems difficult to treat the CDL bubble by the stochastic approach.However, we make an observation that there exists a path integral representation to solutions of the FP equation.In condensed matter physics, it is called Martin-Siggia-Rose-Janssen-de Dominicis (MSRJD) functional integral [16][17][18][19].
In this paper, we use the MSRJD path integral formula to study the tunneling processes on the fixed de Sitter background.The applicability conditions of this method are clarified based on the Schwinger-Keldysh formalism.We consider a potential V (ϕ) which has a local minimum at ϕ = ϕ false , a true minimum at ϕ true , and a local maximum at ϕ top as shown in FIG. 1.We use the saddle point method and evaluate the tunneling rate by computing the action for the tunneling configurations [20].We investigate both HM and CDL tunneling processes.In the case of a shallow potential barrier, the conventional HM tunneling rate is reproduced.Remarkably, the tunneling configuration in the MSRJD functional integral method has physically natural boundary conditions in contrast to the Euclidean method.We then clarify that the HM tunneling rate is the tunneling rate of a (roughly Hubblesized) coarse-grained patch transitioning from ϕ false to ϕ top .In the case of a steep potential barrier, we find an interesting CDL bubble-like configuration which is obtained by solving the solution of a scalar field equation in Euclidean anti-de Sitter space even though we do not work in the imaginary time.Our results show how the bubble nucleation process could be described in the stochastic approach.
This paper is organized as follows.In sec.2, we present the setup and review the stochastic approach.In sec.3, the MSRJD functional integral is constructed from the Langevin equations.We also outline the derivation of MSRJD functional integral starting from the Schwinger-Keldysh formalism.In sec.4, we apply the MSRJD functional integral to the case of the shallow potential barrier and study the HM tunneling process.In sec.5, we consider the case of the steep potential barrier and find the CDL bubble-like configuration.We derive the tunneling rate using the configuration and compare it with the one obtained in sec.4. We also compare our results with those in the Euclidean method.Sec.6 is devoted to the conclusion.Some technical details are presented in appendices.

Stochastic approach
In this section, we review the stochastic approach [7,8].

Setup
We consider a real scalar field ϕ in (3+1) dimensional de Sitter background.The background geometry is described by the metric ds 2 = −dt 2 + a(t) 2 dx 2 = a(η 2 )(−dη 2 + dx 2 ), (2.1) with the scale factor a(t) = e Ht = − 1 Hη , where H is the Hubble constant.The action is given by where a dot denotes a derivative respect to t and V (ϕ) is a potential function.From this action, the canonical conjugate momentum Π is defined as Π := a 3 φ and the Hamilton's equations of motion are given by φ = a −3 Π , Π = a∇ 2 ϕ − a 3 V ′ (ϕ) . (2.4)

Langevin equation
First, we divide the quantum fields ϕ and Π into an infrared (IR) part and an ultraviolet (UV) part respectively as The UV part is defined as where u k is a mode function that will be specified later, and a k and a † k are annihilation and creation operators, respectively.Here, θ(k − k c (t)) is a step function and k c (t) is the cutoff scale defined as where ε is a constant.The value of ε will be discussed later.Substituting Eqs.(2.6) into Hamilton's equations (2.4), we obtain the following equations; where ξ ϕ and ξ Π are defined as To derive Eqs.(2.8) and (2.9), we assumed that the time evolution of UV modes can be well approximated by the free theory that is defined around the false vacuum.The mode function u k (t) introduced in (2.9) then satisfies the following equation From this equation, we see the mode function only depends on k := |k|.Hence, we simply denote the mode function as u k .Note that this mode function is used for UV modes k > k c (t).Let us consider statistical properties of ξ ϕ and ξ Π .The annihilation and creation operators a k and a † k satisfy following commutation relations: Defining the vacuum state |0⟩ as we can calculate the one point and two point correlation functions of ξ ϕ and ξ Π as where Because of the Wick theorem, all higher correlation functions can be decomposed into two point functions.In the stochastic approach, we replace the operator ξ α by the real random field which has the following statistical properties where ⟨• • •⟩ ξ represents an expectation value with the distribution function of ξ α .With this prescription, Eqs.(2.8) can be interpreted as the Langevin equations with the noise ξ α stemming from quantum fluctuations.Since ξ α are now regarded as classical variables, we ignored the imaginary parts of g ϕΠ and g Πϕ which come from non-commutativity.Accordingly, IR variables ϕ IR and Π IR are also regarded as classical stochastic variables in this prescription.The validity of this classical approximation will be discussed in the next section.

Bunch-Davies vacuum
Now we define the vacuum state |0⟩ by specifying the mode function u k .Eq.(2.10) is the equation for the mode k > k c .Let us assume that we can ignore the term V ′′ in solving (2.10).This can be justified when |V ′′ | is sufficiently small to satisfy max (k c /a) where C 1 , C 2 are constants of integration and H (1) ν (z) are Hankel functions.We choose the Bunch-Davies vacuum, i.e., the vacuum coincides with Minkowski vacuum at past infinity (η → −∞) so that (2.17) This determines C 1 , C 2 as Thus, we obtain the mode function To check (2.17), we used the following formulae (z) = 2 πz 3 2 From (2.8), we have to scale g αβ like as g ϕϕ → g ϕϕ , g ΠΠ → g ΠΠ /(H 2 a 6 ) and g ϕΠ → g ϕΠ /(Ha 3 ) in order to compare these quantities.Then, g ϕϕ and g ΠΠ become dominant for the case ε ≪ 1 and ε ≫ 1, respectively.

Reduced Langevin equation
Let us focus on physics at the super-horizon scales by taking ε ≪ 1 and assume a shallow potential so that we can use the massless approximation to evaluate the noise correlations.
In this setup, we can use (2.21) from which we find that g ϕϕ becomes dominant.We then ignore the noise for momentum ξ Π .We also ignore the gradient term in (2.8) because we focus on the physics at super-horizon scales.Since the potential is assumed to be shallow, the time variation of ϕ IR (t) will be then very small.Therefore, we may integrate the second equation of (2.8) as Substituting this relation into the first equation of (2.8), we obtain where the two point correlation function of ξ ϕ is given at the leading order by Eq.(2.23) is the Langevin equation with the noise correlation (2.24) derived in [8].
3 Path integral formalism for the stochastic approach In this section, we introduce path integral approach for the Langevin equation following [21].From now on, we replace the IR field and its conjugate momentum as (ϕ IR , Π IR ) → (ϕ c , Π c ).As we see in section 3.2, this notation corresponds to the Keldysh basis.In section 3.2, we provide a first-principle derivation of the stochastic approach based on the Schwinger-Keldysh formalism following [22,23]. 1 We briefly show the sketch of the derivation in the main text, and details of derivations are given in appendix A.

0+1 dimensional theories
Eq.(2.23) can be regarded as the equation determining the position of a particle ϕ(t) with the white noise ξ ϕ defined by Eq.(2.24).For k c r ≪ 1, we have sin(k c r)/(k c r) ≃ 1 in a good approximation.Let us discretize the field and noise as ϕ c (t) → ϕ i , ξ ϕ (t) → ξ ϕ i , (i ∈ Z).Then, Eq.(2.23) reads where ∆t is the temporal discretization interval.Note that the discretization (3.1) is called Ito discretization. 2Denoting a solution of (2.23) by ϕ c [ξ], the expectation value of an observable ⟨O⟩ ξ can be formally expressed as where Dϕ c := i dϕ i is the functional measure, δ( ), and |δX/δϕ c | is the determinant of {δX i /δϕ j }.When we choose the above discretization (3.1), δX/δϕ c becomes triangular matrix with a unit in the diagonal components, i.e., the functional determinant is unity.Thus, substituting (3.1) into (3.2),we obtain Representing the delta functions in terms of a Fourier integral and taking the continuum limit, we obtain the following path integral representation 1 For earlier discussions of stochastic approach based on the Schwinger-Keldysh formalism, see e.g., [11,24].
2 Generally, we can choose other discretization like as Stratonovich discretization.This ambiguity does not affect the result since the amplitudes of the noise do not depend on fields within the range of our approximation.
After averaging over the noise assuming Gaussian statistics, we obtain the Martin-Siggia-Rose -Janssen-de Dominicis (MSRJD) functional integral ) and taking the boundary conditions ϕ c (t ′ ) = ϕ ′ c and ϕ c (t) = ϕ c , we obtain the transition probability p(ϕ c , t|ϕ ′ c , t ′ ) Finally, introducing new variables Π ∆ := i φ, we obtain the "phase space" path integral where we defined the Hamiltonian as We emphasize that this is just a change of integration variable, not the Wick rotation.A set of saddle point equations is defined as the Hamilton's equations for the Hamiltonian defined above.
In sections 4 and 5, we take the initial state and the final state to be the false vacuum and the field configurations after tunneling (true vacuum or bubble), respectively.

Transition probability from the Schwinger-Keldysh formalism
In the above prescription, we treat IR fields (ϕ IR , Π IR ) as classical stochastic variables.To quantify the validity of this approximation, we derive the stochastic approach from the first principles based on the Schwinger-Keldysh formalism.Note that we recover the label "IR" for IR fields in this subsection following the notation adopted in appendix A.
Suppose that the system is in the false vacuum |Ψ false ⟩ at the past infinity t = −∞ and it evolves into a certain quantum state |{ϕ IR (T, x)} x∈D ⟩ at the final time t = T , where |{ϕ IR (T, x)} x∈D ⟩ is an eigenstate of an IR field φIR (x) at spatial points x in the domain D with an eigenvalue ϕ IR (T, x).Here, φIR (x) is an IR field in the Schrödinger picture. 3The transition probability p for this process is given by where Û (t, t ′ ) describes the unitary time evolution from t ′ to t.The RHS is the Fourier component of the generating functional Z[J IR ] for IR fields, and we have where the generating functional Z[J IR (T )] is defined by We can evaluate Z[J IR (T )] non-perturbatively in the IR sector based on the method developed in [22,23].Our strategy is that we integrate out UV sector k > k c (t) perturbatively in nonlinear couplings to obtain Z[J IR (T )].This is analogous to tracing out environmental degrees of freedom to evaluate the reduced density matrix for the system under considerations.Each step we take can be summarized as follows: 1. First, we derive a path integral representation of Z[J IR (T )].We then split the integration variables into UV fields and IR fields such that the integration contour of UV variables of ϕ is closed: see discussions around Eq.(A.4) for more details. 4.We perform the integration over UV variables and evaluate an IR effective action, the so-called Feynman-Vernon influence functional [25] perturbatively.
3. We substitute the obtained expression of Z[J IR (T )] into (3.13),giving rise to the path integral expression for the transition probability p.
Technical details are shown in appendix A. At the step 2, we assumed that the quantum state for non-zero modes is given by the Bunch-Davies vacuum state for a free field defined around the false vacuum, and that the exact zero mode provides a non-fluctuating classical background field configuration ϕ = ϕ false .
In this way, we can obtain the path integral expression for the probability p as Here, δ [C fin ] and δ [C ini ] fix the boundary conditions of path integral at the final time and the initial time, respectively: ) where δ[C detail ] is not relevant for the main points here, and we define it in appendix A. A term S may be understood as an IR effective action, ) where H is given in (3.10).δH higher denotes the higher-order terms in the coupling constant.Note that eq (3.15) allows us to understand the transition probability p in terms of the stochastic dynamics: this point is discussed in appendix A.3.
Validity of Eq.(3.9).Now we find that S coincides with the term in the exponent of (3.9) up to the term δH.Hence, (3.15) shows that the result (3.9) in the previous section can be justified from the first-principles when the term δH is negligible.Suppose that the potential V (ϕ) is sufficiently flat in the regime of our interest such that perturbation theory works when integrating out UV modes.In this case, we can systematically calculate δH higher perturbatively and we have δH higher = 0 at the leading order.We set δH higher = 0 in the present analysis.We should keep in mind however that the non-perturbative physics at UV scales k > k c (t) = εa(t)H is lost by setting δH higher = 0. From this viewpoint, we should choose ε as large as possible.
Even after setting δH higher = 0, we have nonzero δH.In the classical approximation, we may regard the ∆-variables (ϕ IR ∆ , Π IR ∆ ) as tiny quantities and keep only the terms which are linear in the ∆-variables in S, leading to δH ≃ 0. In the case ε ≪ 1, this approximation would work thanks to the squeezing of quantum fluctuations at super-horizon scales.However, it is more subtle if we can set δH = 0 in the case ε ≳ 1.We will revisit this issue in section 5.

Hawking-Moss tunneling
In this section, we discuss the Hawking-Moss (HM) tunneling from the perspective of MSRJD functional integral.We calculate the tunneling rate defined by (3.7) and (3.9) in sections 4.1 and 4.2, respectively.Hereafter, for the numerical calculations, we consider the potential [26] in FIG.2; for which the derivative with respect to ϕ is given by a simple form For this potential, we have ϕ false = −α, ϕ true = α and ϕ top = −βα.The dimensionless parameter β must be 0 < β < 1, otherwise ϕ = −α is not a false vacuum.The potential at each point takes the following values: 3) The height of potential barrier ∆V between the false vacuum and the true vacuum is The mass of fluctuations around a given point ϕ is In our analysis, we work in the quantum field theory on the fixed de Sitter background.This assumption will be valid when the condition max pl H 2 is satisfied.This is always satisfied when the following condition is imposed, where we use 0 < β < 1 and choose parameters (g, gα/H) = (1.12 × 10 −4 , √ 140) as a benchmark point for later convenience.This shows that we can ignore the backreaction consistently with the weak coupling g ≪ 1 provided that H is well below the Planck scale M pl ≈ 2.44 × 10 18 GeV.Note that under this condition, our discussion below is also applicable in the case of a quasi de Sitter background.However, it is non-trivial to generalize our analysis to the case when a non-negligible backreaction on H exists.This is because it is necessary to extend the first-principles derivation of the stochastic approach to account for the backreaction.This is in itself an interesting issue which we leave for future work.

The case for the reduced Langevin equation
Let us start from the simpler formula (3.7).The Hamilton's equations for the Hamiltonian H(ϕ c , Π ∆ ) defined in (3.8) are given as There exist many Hamiltonian flow lines in phase space (ϕ c , Π ∆ ) that represent the solution of Hamilton's equations as shown in FIG. 3.Each flow line can be specified by the value of Hamiltonian since it is conserved on the given flow line.In the FIG.3, there are two important flow lines corresponding to H(ϕ c , Π ∆ ) = 0. Since the Hamiltonian is the quadratic polynomial of Π ∆ , there are two Π ∆ satisfying H(ϕ c , Π ∆ ) = 0: In the case of Π ∆ = 0, the equations (4.7) reduce to These two flow lines have three intersections which correspond to the false vacuum, the top of potential hill and the true vacuum from left to right.On the flow line with Π ∆ = 0, where t * is the time that ϕ c reaches the top of potential and ∆V is the difference of energy density between the top of the potential and the false vacuum.The trajectory from the top of the potential to the true vacuum does not contribute to the transition probability because H(ϕ c , Π ∆ ) = 0 and Π ∆ = 0 on its trajectory.Thus, the transition probability of the tunneling can be evaluated as This shows a complete agreement with the result of the HM instanton on the fixed background [27].
Comments on this result are in order.It is pointed out in [27] that when the potential V (ϕ) has several degenerate local maxima, all of them yield the same tunneling action in the Euclidean method, despite differing distances from the false vacuum.In our formalism, however, a transition to a distant maximum is described by a saddle-point solution that passes through all the intermediate maxima, and the transition probability is given by the product of the factors on the RHS of (4.12) for each maximum.The aforementioned subtlety is absent in our formalism.
Since the analysis in sec.4.1 is based on (2.23) describing the stochastic dynamics of the IR field ϕ c at a single spatial point, eq.(4.12) calculates the tunneling probability at a single spatial point, not at the whole universe. 5Hence, physically speaking, (4.12) provides the tunneling probability of a coarse-grained patch with a physical radius (εH) −1 .This point becomes manifest also in sec.4.2.

The case for the full Langevin equation
Next, we evaluate the tunneling rate based on the formula (3.9).An advantage of (3.9) is that one can discuss various non-trivial dynamics in the whole universe covered by the flat chart in principle.
In the previous section, we focused on the dynamics at a single spatial point and found the HM tunneling process as a non-trivial solution of the Hamilton's equations.In this section, we discuss the corresponding process in the global picture based on (3.9) by identifying a non-trivial configuration that satisfies the Hamilton's equations in a good approximation.We then reproduce the result (4.11).We also estimate the characteristic time scale of the tunneling.In sec.4.2.1, we remark some technical complications that arise in the global picture.
We start with simplifying the expression (3.9).As discussed in section 2, G ϕϕ becomes dominant and the others can be neglected for ε ≪ 1 in (3.10).Under this approximation, the exponent of the integrand of (3.9) is linear in ϕ ∆ .We can then perform the path integral Dϕ ∆ in (3.9), yielding the product of delta functions x δ Πc −a∇ 2 ϕ c +a 3 V ′ (ϕ c ) .These delta functions give an equation of motion for Π c and eliminate the path integral DΠ c .We solve this equation as Π c ≃ −a 3 V ′ (ϕ c )/3H, ignoring the spatial gradient and the time variation of ϕ c .Furthermore, we change the integration variable as Π ∆ → −iΠ ∆ in (3.9).Consequently, (3.9) reduces to ) where the Hamiltonian H HM is defined as and Π ∆ is given by Here we define t k by the condition k c (t k ) = k.The Hamilton's equations are Here, the term Π∆ (t, x) on the LHS of (4.16b) is defined as the Fourier transform of Π∆ (t, k) i.e., Π∆ (t, x) := (2π) −3 d 3 k Π∆ (t, k)e ik•x .The same rule applies to other IR variables in the real space as in eq.(A.25).Intuitively, this is because the stochastic formalism is formulated in the momentum space and its Fourier transform provides the stochastic formalism in the real-space: see appendix A for more details.Eqs.(4.16) admit a trivial configuration Π ∆ (t, k) = 0 under which we have H HM = 0 and (4.16a) becomes φc = −V ′ (ϕ c )/(3H) describing the standard classical time evolution.Now we are interested in the non-trivial configuration which satisfies (4.16).In the previous section, we focus on the non-trivial dynamics at the single spatial point x = x 0 and it is found that the tunneling configuration is obtained by Π ∆ (t, x 0 ) = − 8π 2 3H 4 V ′ (ϕ c (t, x 0 )).This configuration can be naturally extended to the one in the global picture by considering the following configuration From now on, we show that this configuration satisfies a set of Hamilton equations (4.16) in a good approximation.For this configuration, we have where r := |x − x 0 | and f (r; t) := (2π) . The functions j 0 (k c (t)r) and f (r; t) rapidly oscillate for r ≫ k −1 c (t).Hence, they exponentially decay to zero after they are smeared over the Hubble time: for instance, j 0 and f becomes the smooth function after the suitable coarse-graining as . (4.20) Here, we perform the smearing by using a window function w(t ′ , t) which is approximately constant in the domain |t−t ′ | ≪ H −1 while it exponentially decays to zero at |t−t ′ | ≫ H −1 .
We also impose the normalization condition dt ′ w(t ′ , t) = 1.We can choose w(t ′ , t) = Hπ −1/2 exp[−H 2 (t − t ′ ) 2 ] for example.Note that the property (4.20) is based on the behavior j 0 (k c r) ≃ 1 and f (r; t) and the rapid oscillations of j 0 (k c r) and f 0 (r; t) for r ≫ k −1 c .Hence, the property (4.20) will be robust against the choice of w(t ′ , t).

.21b)
This exhibits the non-trivial dynamics only in the domain r ≲ k −1 c , i.e., a coarse-grained patch centered at a point x 0 with a physical radius (εH) −1 .Here, we used ϕ c (t, x) ≃ ϕ c (t, x 0 ) for r ≪ k −1 c (t).In the second line, it is assumed that V ′ (ϕ c (t, x 0 )) does not cancel the rapid oscillations of j 0 (k c (t)r) so that their product V ′ j 0 decays exponentially at r ≫ k −1 c after the suitable smearing.This assumption will be valid thanks to the approximate constancy of super-Horizon fluctuations ϕ c (t, k) over the Hubble time: Adopting the same assumption, from (4.18a) and (4.18b) we also obtain the smeared expressions of Π ∆ and Π∆ which turn out to satisfy (4.16b) under the condition (4.16a).Hence, we conclude that (4.18) describes a non-trivial saddle-point solution in a good approximation.Furthermore, after the smearing we can find H HM ≈ 0 for this configuration.Now we use the configuration (4.17) or (4.18) to obtain the tunneling configuration shown in FIG. 4 again.We can calculate the action for this tunneling configuration similarly to (4.11).By substituting the exact expressions (4.18) and (4.21a), we obtain 7 In the first line, we used (4.16a) to eliminate φc from the integrand.In the second line, we substituted the concrete configurations (4.18).In the third line, we used (4.21a) and performed the spatial integral as d 3 x j 0 (k c (t)r) f (r; t) = 1.Thus, we reproduce the previous result (4.11).Our analysis confirms that (4.22) gives the tunneling probability of a coarse-grained patch with a physical radius (εH) −1 .
It is worth mentioning the time scale of the HM tunneling.We may naively calculate the typical time scale of the HM tunneling as t * − t ′ = ϕtop ϕ false dϕ c 3H V ′ (ϕc) .Here, the spacetime argument of the field is suppressed.However, this integral is in general divergent since V ′ = 0 at the both ends of the integral.In the vicinity of the domain where V ′ = 0, we expect that the quantum fluctuations of the field would play an important role.Since the typical size of quantum fluctuations accumulated over the Hubble time is H/(2π), we 7 Note that we can also estimate the action I by performing the coarse-graining in time suitably and substituting the smeared expression of Π∆ into the action: reproducing the previous result (4.11).In the first approximate equality, we used HHM = 0 which is well satisfied after the smearing.In the second one, we use ϕc(t, x) ≃ ϕc(t, x0) for r ≪ k −1 c (t) and the condition d 3 x W f (r; t) = 1 which follows from the definition of W f .This estimate may be more analogous to the calculation of the action in sec.4.1 since the approximate locality of the dynamics is restored after the smearing.
may define the regulated quantity t HM as and we expect that t HM would correctly characterize the typical time scale of the HM tunneling.In particular, for our potential (4.1) we have (4.24) Since we have (H/gα) 2 ∼ (H 2 /V ′′ ) ϕ false ≲ϕc≲ϕtop ≫ 1 for the shallow potential, we conclude that the time scale of HM tunneling is much longer than the Hubble time.

Remark
So far it is found that, starting from (3.9), the configuration (4.17) approximately solves the Hamilton's equations (4.16) and reproduce the previous result (4.11) when it is substituted into the action.There is actually an important reason behind why we should choose the configuration (4.17) to evaluate the HM tunneling process.
To see this, it is important to realize an important difference between the current Hamilton's equations (4.16) and the previous one (4.7); the equation for ϕ c now contains Π ∆ rather than Π ∆ .Due to this difference, the structures of Hamilton's equations (4.16) are understood as follows; • the equation for ϕ c (4.16a) is determined once Π ∆ is specified, • Π ∆ (t, x) is understood as initial conditions for (4.16b) since Π ∆ (t, x) contains only the boundary modes satisfying k = k c (t).Hence, Π ∆ (t, x) is obtained by solving (4.16b) for given Π ∆ under the condition (4.16a).
On top of that, substituting the Hamilton's equation for ϕ c into the action, we have the first line of (4.22) which can be rewritten as The RHS depends only on the value of Π ∆ (t, k)| k=kc(t) .Hence, we can evaluate the tunneling action once Π ∆ is specified, provided that (4.25) or equivalently (3.9) is reliable.We claim that (3.9) will be valid when Π ∆ is chosen so that Π ∆ (t > t k , k) becomes a smooth function in time.Otherwise, it will not be justified to ignore the higher-order corrections δH higher which are neglected in (3.9).Intuitively, this states that perturbation theory tend to be broken down around exotic configurations.When Π ∆ is a smooth function, we can relate Π ∆ to Π ∆ after the suitable smearing in time: Here, we also used Π We then require that the Hamilton's equation for Π ∆ (4.16b) is satisfied in a good approximation to ensure the validity of (3.9).This constrains the choice of Π ∆ .Indeed, the configuration (4.17) approximately solves (4.16b) and satisfies the smoothness and henceforth (4.26) after the smearing.This would be the reason why our analysis with (4.17) based on (3.9) can reproduce the correct result.

On the ε-independence of the Hawking-Moss tunneling
In sections 4.1 and 4.2, the tunneling probability of a coarse-grained patch with a physical radius (εH) −1 is calculated.Our results coincide with the HM tunneling and are independent of ε.
The ε-independence of the results would be a consequence of the scale-independence of the dynamics of light scalar fields at super-horizon scales.To see this, it is useful to notice that a coarse-grained patch of physical radius (ε 1 H) −1 at a time t = T expands to a patch of larger physical radius (ε 2 H) −1 at a later time t = T + δt > T with The value of IR field does not evolve from t = T to T + δt because the fluctuations of light scalar field at super-horizon scales are approximately time-independent, unless δt is much longer than H −1 .Hence, we can relate the coarse-grained dynamics of different ε ≪ 1 by considering the time shift t → t + δt with the value of IR field being kept fixed.This means that the tunneling probability from ϕ = ϕ false to ϕ true of a patch with physical radius (εH) −1 should be independent of ε.

Coleman-de Luccia tunneling
In the previous section, the result of the HM instanton is reproduced by using (3.9).We expect that the formula (3.9) can describe not only the HM tunneling but also the CDL tunneling.In other words, we expect that there will be a flow line starting from the false vacuum to the bubble configurations as illustrated in FIG. 5.In this section, we concretely show an interesting configuration by following the prescription in the previous section.

Coleman-de Luccia bubble solution
In Euclidean method, the CDL instanton is dominant rather than the HM instanton when the potential barrier is steep [28][29][30][31].By taking ε ≫ 1, we can study nonperturbative physics at sub-horizon scales such as the formation of bubble.In this case, G ΠΠ becomes dominant and the other noises can be ignored in (3.10).Under this approximation, the exponent of the integrand of (3.9) is linear in Π ∆ .We can then perform the path integral DΠ ∆ in (3.9), yielding the product of delta functions x δ φc − a −3 Π c .These delta functions eliminate the path integral DΠ c . 8Furthermore, we change the integration Figure 5: A schematic picture of Hamiltonian flows in the "field" phase space.We expect that there is a flow line which starts from the false vacuum (left intersection) and reaches the bubble configuration (right intersection) through a non-zero ϕ ∆ (x) path.variable as ϕ ∆ → −iϕ ∆ in (3.9), leading to with appropriate boundary conditions; we consider bubble configurations as boundary conditions later.Here, the Hamiltonian H CDL is defined by with We would like to find the appropriate configuration ϕ ∆ which describes the tunneling process.Following the choice made in sec.4, let us suppose that the appropriate choice is given by the configuration for which the Hamiltonian (5.2) vanishes: Note that this condition is imposed up to the suitable smearing in time because ϕ ∆ (t, x) includes only the boundary Fourier mode k = k c (t).As discussed in sec.4.2.1, ϕ ∆ becomes sufficiently smooth for the appropriate ϕ ∆ , and hence we also suppose after the suitable averaging over the Hubble time, similarly to (4.26).We use the relation (5.5) to evaluate the action later.This would work at least for the purpose of estimating the action.
In principle, we do not need to use the estimate (5.5) in evaluating the action since it is determined once ϕ ∆ is specified based on the analogous logic discussed around (4.25).For this purpose, we need to specify the configurations {ϕ ∆ (t, k)| k=kc(t) } in momentum space which result in the sufficiently smooth ϕ ∆ that solves the Hamilton equations.It is not easy to find such appropriate configurations precisely, however.Hence, we leave more careful analysis on the choice of ϕ ∆ for future work.In this study, instead, we focus on how we can proceed the analysis for given configuration (5.4) with adopting the estimate (5.5) and how the bubble nucleation process could be described in the stochastic approach.
Using the relation (5.5), the Hamiltonian (5.2) and the configuration (5.4) become ) (5.7) The Hamilton's equations under the constraints (5.7) give the following equation of motions for ϕ c : Similar to the previous case (4.9), the signs of the potential term and the gradient term are flipped in the equation of motion with the non-trivial ϕ ∆ ̸ = 0. Hence, we can expect that the tunneling process is realized.From (5.1), we find that only the solutions satisfying ϕ ∆ ̸ = 0 contribute to the action.Then we focus on the second equation in (5.8).
From the global coordinate (5.11), we see the Euclidean AdS spacetime has O(4) symmetry.Hence, we assume that the field ϕ c depends only on ρ.Thus the second equation of (5.8) reads Note that this is the same equation utilized in the Euclidean method [3,5].Imposing the boundary conditions [1,3] lim we obtain the bubble solutions as is shown in FIG. 6.Note that the latter condition ensures the continuity of ∂ η ϕ c (η, r) and ∂ 2 η ϕ c (η, r) at (Hη, Hr) = (−1, 0).It is useful to fit the bubble configurations by the following fitting function (FIG.6); where µ and ρ represent the thickness and the position of the bubble wall, respectively.Note that, for β ≳ 0.6, the deviation from the true vacuum at the ρ = 0 becomes significant [26] and the fitting (5.17) becomes bad.We also have to examine the boundary conditions in the Hamilton flow.As in the case of HM tunneling, we expect that there are several intersections where ϕ ∆ = 0 (FIG.5).Since we have the concrete ϕ ∆ (5.7) and the bubble configurations (5.17), it is possible to obtain the curves in (r, η)-plane on which ϕ ∆ = 0. Using (5.7) and (5.8), such curves are given as Note that this is equivalent to solve a −2 ∇ 2 ϕ c − V ′ (ϕ c ) = 0. Substituting (5.7) into (5.18) and using the fitting formula (5.17), we find four solutions which are expressed as four lines in the (η, r)-plane; two of them are placed at η = −∞ with fixed r and r = ∞ with fixed η.These correspond to ρ = ∞ where ϕ = −α (FIG.6).Also, we numerically find other two non-trivial hypersurfaces (FIG.7(a)).It can be seen that one of the curves is totally spacelike but another is partially timelike.Thus, it seems natural to take η = −∞ curve as an initial time slice on which ϕ is false vacuum and totally spacelike non-trivial curve as a final time slice on which the bubble is nucleated.We denote the region between the two spacelike curves as Σ.
As a matter of fact, the configurations on the final time slice has a bubble.It can be checked as follows; the value of ϕ c at the origin exceeds the top of potential hill.In the case of FIG. 7  We use the fitting function (5.17) for the numerical integration.
In the second line, we used H CDL = 0, (5.7) and (5.8).Also we performed a coordinate transformation (t, r) → (ρ, θ 1 ).In the third line, the following relations are used and Here, the partial differentiation with respect to η is taken while keeping r constant.

Appropriate choice of ε
Since our results (5.19) depend on ε, we need to choose some specific value of ε to predict the tunneling rate.We emphasize that the ε dependence does not imply the pathology of our result.Rather our results should depend on ε because the non-perturbative effects from UV modes are missed in our formalism as discussed in section 3.2.We then choose ε as large as possible to evaluate the tunneling rate.Our analysis will be valid only when the value of ε lies in a certain range.Below, we briefly discuss necessary conditions for ε to justify our analysis.
Lower bound on ε.We need to choose sufficiently large ε so that IR fields can describe the bubble configuration.This imposes that (εH) −1 should be smaller than the typical physical size of the bubble or the bubble wall.This condition will be satisfied by taking ε ≳ µ for our parameter choices.
Upper bound on ε.Our analysis is based on Eq.(3.9) which will be invalid when quantum fluctuations of IR fields are too large.This imposes an upper bound on ε because the size of fluctuations of ϕ c at the scale k ∼ k c (t) is proportional to ε: where we assumed (k c /a) 2 ≫ V ′′ (ϕ c ) 9 and approximated ϕ as a massless scalar field.As discussed in section 3.2, (3.9) will be justified when the term δH is negligible.Then, we define the ratio Q of the term δH to the final term on the RHS of (5.6) as10 and we impose Q < 1.In the second equality, we used δH = (a 3 /24)V ′′′ (ϕ c )ϕ 3 ∆ .Similarly to (5.20), we can estimate the size of quantum fluctuations of a −2 ∇ 2 ϕ.Then, we estimate Q at large ε as Q ∼ g 2 αε/H where we also used Summary.In summary, it is necessary to choose ε satisfying the following condition to justify our analysis, .21)This suggests that, at least when g is sufficiently tiny, we can reliably choose large ε.Note that we cannot make the action I given in (5.19) arbitrarily small even if we take ε max → ∞ in the free-theory limit g → 0; for ε = ε max , we have I| ε=εmax ∼ −g −1 (gα/H) 3 Ĩ(µ, ρ).Roughly speaking, the small gα/H corresponds to the bubble radius being larger than the Hubble radius.Since we are interested in the bubble smaller than the Hubble radius, gα/H cannot be taken small arbitrarily.
A few more comments on the ε-dependence of our results are in order.Our result (5.19) shows that the value of I becomes larger for smaller choice of ε.This is because the size of quantum fluctuations of modes k ∼ εa(t)H is proportional to ε, while the nonperturbative physics caused by UV modes k > εa(t)H is truncated in our formalism.This interpretation however implicitly assumes that the modes k ∼ ε max a(t)H are relevant for the tunneling process.We expect that this is the case because a kinetic energy of fluctuations of modes k ∼ ε max a(t)H is estimated as (ε max H) 4 which is much smaller than the height of potential barrier ∆V ∼ g 2 α 4 for our parameter choices.
By contrast, very short-scale physics which is insensitive to the detailed structure of V (ϕ) near its origin might be irrelevant for the tunneling process.Hence, we expect that the tunneling rate would converge to some finite value in the limit ε → ∞.It would be interesting to see whether the tunneling rate converges to a finite value as ε increases, and if so, from which value of ε this convergence begins.

Hawking-Moss vs. Coleman-de Luccia
From the formula (3.9), we derived the two configurations which describe the tunneling from the false vacuum to the true vacuum.One is HM configuration and the other is CDLlike configuration.Using these configurations, we evaluate the probability of the tunneling.The values of γ for several parameter sets.First and second ones are potential parameters.Third one is the initial condition for the equation of motion (5.15).Forth and fifth ones are fitting parameters for (5.17).Sixth and seventh ones are the numerical results of (5.22) and (B.9).The last one is the cut-off parameter ϵ for which (5.22) coincides with (B.9).
Here, we compare these probabilities.Denoting the action (5.19) and (4.11) as I bubble and I HM , the ratio of the two actions is given by For example, if we choose β = 0.5 and αg/H = √ 140 (see FIG. 6), the fitting parameters become µ = 7.32515 and ρ = 0.34397 and the ratio becomes γ = 12.9840/ε.For these parameters, the upper bound of ϵ is given as (5.21) and then γ can be smaller than one for appropriately small g.This means that the CDL bubble configuration is dominant rather than HM configuration.Results for other parameters are also shown in Table 1.

Comparison with the Euclidean method
We also compute the CDL action in Euclidean method (Appendix B) and compare the ratios of the CDL action to the HM action (Table 1).For ε > ε * , our tunneling process becomes more probable than the one predicted by the Euclidean method.As discussed in section 5.2.1, we can consistently choose ε > ε * for certain choices of parameters (g, α, β).Perhaps, our results may indicate the presence of tunneling process which is more probable than the Euclidean method.

Bubble nucleation hypersurface and the subsequent evolution
We find the non-trivial spacelike hypersurface on which the bubble is nucleated (FIG.7(a)), where the hypersurface is given by the condition F (η, r) = 0 or equivalently a −2 ∇ 2 ϕ c − V ′ (ϕ c ) = 0.The subsequent evolution is then described by the one shown in the first line of (5.8) which is the standard classical equation of motion.The configuration on the hypersurface gives the initial data for the classical dynamics after the bubble nucleation.
The field value on the spacelike hypersurface does not reach the true vacuum but rather it lies between the true vacuum and the top of the potential as mentioned in the subsection 5.1.This is possible because the location of hypersurface is defined by the condition a −2 ∇ 2 ϕ c − V ′ (ϕ c ) = 0 where the gradient force is balanced with the potential force.Though it is non-trivial to solve the evolution starting from the general spacelike hypersurfaces, it is desirable to solve it to fully understand the formation of the true-vacuum bubble in our scenario.
We conclude this section by pointing out that our formalism naturally predicts that the condition ϕ ∆ = 0 (or ϕ ∆ = 0 via (5.5)) defines the hypersurface on which the quantum dynamics is switched to the classical dynamics and the non-trivial field configuration is nucleated.This would be the generic prediction of our formalism that holds true once the appropriate configuration ϕ ∆ is specified.

Conclusion
We studied the tunneling processes on de Sitter background by using the stochastic approach.A novel point is that we applied the MSRJD functional integral to the problem.In this formalism, the tunneling rate is obtained by evaluating the functional integral with the saddle-point approximation.Using this method, we investigated both the HM and the CDL tunnelings.
For the HM case, we first analyzed (3.7) which is the MSRJD functional integral of the stochastic equation (2.23) on a single spatial point.We succeeded in deriving the tunneling solution in the "phase space" (FIG.3) which represents the tunneling process from the false vacuum, through the top of the potential hill, and finally to the true vacuum (FIG.4).This solution has natural tunneling boundary conditions which cannot be obtained from the Euclidean method.The tunneling rate for this configuration coincides with that of the HM instanton at the leading order.In sec.4.2, we succeeded in re-deriving this result starting from (3.9) which describes the stochastic dynamics in the global region covered by the flat chart.We also estimated the time scale of the HM tunneling as (4.23) which becomes much longer than the Hubble time scale.Our analysis clarifies the physical picture of the HM instanton, i.e., the HM transition probability represents the transition probability of a coarse-grained patch with a physical radius (εH) −1 .Some technical complications that arise in dealing with (3.9) and the way of handling them are remarked in sec.4.2.1.Physically, these complications are due to the spatial correlations of stochastic noises.
For the CDL case, we found the configurations which describe the bubble nucleation process.The tunneling rates for the bubble configurations depend on the cutoff scale dividing IR and UV fields.We argued that this dependence comes from the truncation of the non-perturbative effects from UV modes.We also discussed the valid choice of the cutoff scale for which the MSRJD functional integral will be reliable.This consideration is based on the first-principles derivation of stochastic approach from the Schwinger-Keldysh formalism.With an appropriate cutoff scale, it turned out that the CDL tunneling rate is larger than the HM one for the steep potential barrier.However, as mentioned above (5.6), it has not yet been investigated if our configuration can be really used for estimating the tunneling action.This aspect would be important for evaluating the tunneling action, being left for future work.Nonetheless, we believe that our study clarifies how we can proceed the analysis and how the bubble nucleation process could be described in the stochastic approach; for instance, our formalism can naturally define the location of hypersurface on which the quantum dynamics is switched to the classical dynamics and the non-trivial field configuration is nucleated.
It is interesting to understand the relation between the stochastic approach and the Euclidean method.We compared the CDL tunneling rate in our method with that in Euclidean method and found the former becomes larger than the latter for the potential considered with a certain cutoff scale.It would be worth investigating the meaning of this result.Especially, we need to know a precise relation between the CDL instanton and our configuration.Intriguingly, our method is also related to that discussed in [32][33][34][35][36], where the bubble accidentally appears from the initial quantum fluctuations.In our method, however, we assume that we can neglect the non-perturbative effects from UV modes, the quantum component of the potential of the form V ′′′ (ϕ c )ϕ 3  ∆ , and that the configuration (5.7) can be used to estimate the tunneling rate.The first two are summarized into the term δH.We also assume the fixed background spacetime.The relaxation of these assumptions would be important to seek the relations among the three methods.It would also be useful to apply our path-integral method to the tunneling in flat space and make a comparison with the Euclidean method.
Once this relaxation is achieved, it would also be interesting to apply our method to tunneling phenomena where the backreaction to the background geometry is non-negligible and make comparisons with recently-discussed methods.For instance, the formalism of the Wheeler-DeWitt equation [37][38][39] and the tunneling potential [40] must be related to ours because the HM exponent is reproduced.The tunneling in the black hole spacetime [41,42] is also interesting.Especially, [42] discussed the HM transition with a black hole from the viewpoint of the stochastic approach.
Our method using the saddle-point approximation effectively amounts to solving realtime quantum dynamics, which reduces to Starobinsky's stochastic approach [7,8] at leading order when the super-horizon dynamics in de Sitter space is considered.However, since our starting point is the Schwinger-Keldysh path integral, which can be formulated in a more generic setup and incorporates all the quantum effects in principle, we believe that our formalism sheds light on further studies of tunneling phenomena from a real-time perspective.
One of the advantage of our method is that it is applicable to a dynamical setup.The application of our method to the inflation models such as the chain inflation [43] and the warm inflation [44] is also intriguing.Using our method, we can study the tunneling process in the inflationary background under dissipation and fluctuations coming from circumference.We leave these issues for future work.

Funding:
T. M. was supported by JST SPRING, Grant Number JPMJSP2148 and JSPS KAKENHI Grant Number JP23KJ1543.J. S. was in part supported by JSPS KAKENHI Grant Numbers JP17H02894, JP17K18778, JP20H01902, JP22H01220.J.T. is supported by IBS under the project code, IBS-R018-D1.This expression provides a stochastic interpretation of the transition probability; the dynamics of ϕ IR c is described by the set of Langevin equations and the transition probability is simply understood as the probability to realize the field configuration ϕ IR c (T, x) = ϕ IR (T, x) in the domain x ∈ D at the final time T starting from the initial configuration ϕ IR c (−∞, x) = ϕ false .In particular, the set of Langevin equations is given by (2.8) with the noise correlations (2.15) when we ignore the terms δH higher , S IR H,(s) , and Γ k=0 .Note that such terms simply correct the Langevin equations 14 and do not invalidate the above stochastic interpretation.

B Coleman -de Luccia tunneling in Euclidean method
In this appendix, we provide the CDL action in de Sitter background with Euclidean method.
Generally, the ways of Euclideanization for de Sitter spacetime depend on the coordinates.In this paper, we discuss the field theory on the metric (2.1), and then we choose the following coordinate again 15 .ds 2 = 1 H 2 η 2 (−dη 2 + dx 2 ). (B.1) By taking the wick rotation as η → −iτ , the coordinate becomes This is the Euclidean AdS space discussed in section 5 and we can take the global coordinate; −ds 2 = H −2 (dρ 2 + sinh 2 ρdΩ 2 ).(B.3) 14 For instance, leading-order corrections to the stochastic dynamics at the super-horizon scales are calculated in [22]. 15The analytic continuations of this coordinate is examined in [5] We also impose that the field ϕ only depends on ρ.Therefore, the Euclidean action I E and the equation of motion becomes The second equation is the same as (5.15) and we can obtain the same bounce solutions discussed in section 5.For numerical calculations, we consider the potential (4.1) and fitting function (5.17).Also we take the following non-dimensionalizations; ϕ = α φ, (B.6) Therefore, the Euclidean action (B.4) becomes

Figure 1 :
Figure 1: The schematic picture of the potential V (ϕ) which has a false vacuum at ϕ = ϕ false , true vacuum at ϕ = ϕ true and the top of the potential at ϕ = ϕ top .

Figure 3 :
Figure 3: The Hamilton flow for (4.7)where we take α = H, g = 0.4 and β = 0.1.The solid lines correspond to H(ϕ c , Π ∆ ) = 0. From left to right, the intersections of the solid lines correspond to the false vacuum, the top of potential hill and the true vacuum.

Figure 7 :
Figure 7: (a): The two non-trivial curves on which (5.18) is satisfied.We insert the mesh to confirm whether the curves are spacelike or timelike.It can be seen that the upper line is spacelike but the lower one is partially timelike.(b): The integration region for (5.19).There are another spacelike curve at η = −∞.(c): The two non-trivial curves in (ρ, θ 1 ) plane is plotted.(d): The integration region for (5.19) in (ρ, θ 1 ) plane.All plots are the case for β = 0.5 and αg/H = √ 140.

Table 1 :
The ratio of the Coleman-de Luccia Euclidean action I E bounce to the Hawking-Moss action I HM is given as bounce (1 − β) 3 (β + 3) .(B.9)