betaFeedback

WorksPhysics and Astronomy

Nonperturbative lattice simulation of the nucleation rate in thermal first-order phase transitions

Abstract

We present a nonperturbative lattice study of vacuum decay through bubble nucleation in a three-dimensional real scalar field theory at finite temperature. The vacuum decay rate is computed by combining an expanded-ensemble multicanonical hybrid Monte Carlo approach, which samples the critical bubble configurations, with real-time Langevin dynamics, which follows the subsequent bubble evolution. We extrapolate the nucleation rate to the continuum limit and compare it with the corresponding one-loop perturbative prediction, finding a discrepancy of about 8%8\%.

Introduction

The first-order phase transition at the electroweak scale is of great interest in early-universe cosmology, since if such a transition exists in extensions of the Standard Model, it may provide the necessary out-of-equilibrium conditions for electroweak baryogenesis [1, 2, 3, 4, 5, 6, 7], give rise to the cosmic origin of primordial magnetic fields [8, 9, 10, 11, 12, 13, 14, 15, 16, 17], produce primordial black holes [18, 19, 20, 21, 22, 23], and generate primordial gravitational waves observable by space-based gravitational-wave detectors [24, 25, 26, 27]. Moreover, several Pulsar Timing Array collaborations have reported evidence for a stochastic gravitational-wave background [28, 29, 30, 31, 32, 33], opening the possibility that new-physics phase transitions contribute to the observed signal. The bubble nucleation rate Γ\Gamma is a central input for characterizing the phase transition dynamics. Because the nucleation rate depends exponentially on the height of the phase transition barrier, theoretical predictions require the error in ∣log⁡Γ∣|\log\Gamma| to be kept under strict control; otherwise, quantitative predictions of the gravitational-wave spectrum are not feasible.

Theoretically, the calculation of the nucleation rate originates from the saddle-point approximation for zero-temperature quantum tunneling [34, 35, 36] and the statistical theory of finite-temperature classical nucleation [37, 38, 39, 40]. In the latter framework, the nucleation rate is factorized into a statistical prefactor and a dynamical factor: the former yields the equilibrium probability of the critical bubble relative to the metastable vacuum, while the latter governs the rate at which the critical bubble traverses the barrier top. This formalism was subsequently extended to finite-temperature quantum field theory [41, 42]. At high temperature, the thermodynamics of a four-dimensional quantum field theory reduces to a three-dimensional effective field theory [43, 44], thereby recasting the nucleation rate calculation as a problem in quasi-classical statistical field theory.

However, the perturbative expansion exhibits poor convergence at higher orders; in particular, the statistical prefactor is dominated by exponentially suppressed field configurations, rendering the calculation challenging to control. The theoretical uncertainties inherent in perturbative approaches have been systematically investigated across a wide range of models [45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55], and the choice of the effective potential and the renormalization scale can further introduce additional scheme dependence [56, 57, 58, 59, 60]. These findings underscore the necessity of nonperturbative calculations for obtaining a reliable nucleation rate.

Lattice methods provide a systematic nonperturbative approach to the statistical prefactor. For weakly metastable systems, direct simulation of the thermal nucleation process in real time has been widely studied [61, 62, 63, 64, 65, 66, 68] [67]. An alternative approach is the multicanonical Monte Carlo method [69, 70], which enables a fully nonperturbative calculation of an extremely small electroweak bubble nucleation rate [71]. Nevertheless, the multicanonical method faces a sampling obstacle known as the condensation barrier, which is difficult to circumvent in large-volume simulations and leads to an exponential decrease in simulation efficiency with volume [72, 73].

Recently, a novel expanded-ensemble method was proposed to effectively circumvent the condensation barrier [74]. This approach grows a bubble through a chain of subvolumes, initiating from a small region where the barrier is negligible and progressively enlarging the subvolume until the subcritical size is attained, thereby ensuring that the tunneling time no longer exhibits exponential growth with volume. Although this method has been validated in a two-dimensional scalar field model with a symmetric double-well potential, its extension to three dimensions remains unexplored, as the form of the potential, the efficiency of the update algorithm, and the continuum-limit behavior all warrant further investigation in the three-dimensional case.

In this work, we extend the expanded-ensemble multicanonical framework to a three-dimensional asymmetric double-well system, enabling the computation of the statistical prefactor and the extraction of the critical bubble configuration. We employ multicanonical hybrid Monte Carlo (HMC) updates to significantly improve sampling efficiency [75, 76]. Crucially, we introduce a translation-invariant, differentiable weighted order parameter OinvO_{\mathrm{inv}} that resolves the fixed-window pinning issue and obviates the need for auxiliary shift updates [74], all while maintaining detailed balance and differentiability for HMC. Coupled with real-time Langevin dynamics, this allows us to determine the complete nonperturbative nucleation rate and benchmark it against the one-loop Langer result. To our knowledge, this constitutes the first successful extension of the multicanonical expanded-ensemble method to three dimensions.

The structure of this paper is as follows. Section II outlines the numerical methodology, while Section III details the lattice implementation. In Section IV, we present the complete numerical results and systematically compare the nonperturbative nucleation rate with the one-loop perturbative prediction. Finally, Section V summarizes our findings and discusses their broader implications.

Numerical Method

In this section, we present the methodology combining the expanded ensemble multicanonical Monte Carlo method with the real-time Langevin technique. The Monte Carlo approach is utilized to extract the statistical prefactor and identify critical bubble configurations. Building upon these configurations, the real-time Langevin dynamics is then applied to evaluate the dynamical prefactor.

Order parameter

To distinguish between the two phases (symmetric and broken) and identify critical bubbles, an order parameter is needed. The simplest choice is the volume average of the field,

O≡1V∑x⃗ϕ(x⃗),(1)\mathcal{O} \equiv\frac{1}{V} \sum_{\vec{x}} \phi(\vec{x}), \tag*{(1)}

or its square ϕˉ2\bar{\phi}^{2}, with the latter being more common in current simulation studies. In this paper, we use an improved quadratic-type order parameter,

O=1V∑x⃗(ϕ(x⃗)−A)∣ϕ(x⃗)−A∣,(2)\mathcal{O} = \frac{1}{V} \sum_{\vec{x}} \left(\phi(\vec{x}) - A\right) \lvert\phi(\vec{x}) - A\rvert, \tag*{(2)}

to be specified bellow. This order parameter not only makes the critical region more prominent, thereby facilitating sampling, but also naturally captures the evolution of the field value, rendering it highly suitable for the HMC updating scheme.

In large-volume simulations, the shift in the order parameter induced by a small bubble can be entirely overwhelmed by bulk fluctuations, rendering the bubble branch in the histogram unresolvable. To address this issue, we introduce a spatial smoothing technique, which replaces the field value with a weighted average

of itself and its nearest neighbors, thereby filtering out short-wavelength fluctuations:

ϕ(n+1)(x⃗)=14(ϕ(n)(x⃗)+12∑μ^ϕ(n)(x⃗+μ^)),(3)\phi^{(n+1)}(\vec{x}) = \frac{1}{4}\left(\phi^{(n)}(\vec{x}) + \frac{1}{2}\sum_{\hat{\mu}}\phi^{(n)}(\vec{x}+\hat{\mu})\right), \tag*{(3)}

where μ^\hat{\mu} runs over positive spatial directions. After nn smearing steps, the smoothed field is denoted by ϕs(x⃗)\phi_s(\vec{x}). Smearing strongly narrows the width of the order-parameter distribution in the pure bulk phase, making the bubble signal more visible at large volumes.

Condensation barrier

Near the condensation transition point, the same value of the order parameter can correspond to two completely different types of field configurations, referred to as the bubble branch and the fluctuation branch, respectively. Figure 1 illustrates this situation. For the same value of the order parameter, the left panel shows a configuration containing a bubble, whereas the right panel shows a configuration with no bubble, thereby illustrating the distinction between the bubble and fluctuation branches. There is a hidden barrier for transitions between these two branches, as the multicanonical weight depends only on the order parameter and cannot distinguish between them. As the system volume increases, the physical size of the critical bubble increases, and the height of the condensation barrier increases as well. Periodic boundary conditions can also introduce shape transitions [77], further increasing the difficulty of sampling.

Examples of configurations with approximately the same value of the volume-averaged order parameter

Figure 1. Examples of configurations with approximately the same value of the volume-averaged order parameter. The left panel shows a configuration containing a small bubble (the bubble branch), while the right panel shows a pure bulk configuration (the fluctuation branch).

Expanded ensemble method

The general multicanonical framework is reviewed in Appendix C; here we directly present the expanded-ensemble construction used in this work.

The expanded ensemble method [74] introduces the weighted order parameter and constructs a chain of subensembles, effectively decomposing the rare event in a large volume into a sequence of easily sampled events in small volumes to overcome the condensation barrier.

Weighted order parameter

The conventional order parameter O\mathcal{O} is measured over the entire system volume VV (e.g., the full-volume O\mathcal{O}), while the weighted order parameter OW\mathcal{O}^{W} is measured only within a specified subvolume VW⊂VV^{W} \subset V, e.g., inside a spherical region of radius rr. When the subvolume VWV^{W} is small, the condensation barrier for forming a small bubble inside it is also small. Therefore, bubbles can be easily created and destroyed with little condensation barrier, and then the system can accommodate configurations containing larger and larger bubbles by gradually increasing the subvolume.

Ensemble chain construction

The entire sampling process is decomposed into a sequence of subensembles i=1,2,…,Ni = 1, 2, \ldots, N, each using a different weighted volume ViWV_{i}^{W} and a carefully chosen order-parameter range. The first subensemble uses the smallest subvolume V1WV_{1}^{W}, whose order-parameter range extends from the metastable peak OmO_{m} to the size of the largest bubble that fits inside this small volume. In this small volume, the condensation barrier can be negligible, allowing the field to fluctuate between no-bubble and small-bubble states freely. In the intermediate subensembles, the subvolume ViWV_{i}^{W} is increased step by step, while the order-parameter range of each subensemble is chosen to overlap with those of the neighboring ensembles, ensuring a continuous chain. These subensembles are designed to contain bubbles and prevent their complete evaporation. The final subensemble uses the full volume VNW=VV_{N}^{W} = V, and its order-parameter range is a portion of that of the previous subensemble. Subsequently, the standard multicanonical method can be directly used in the full volume to explore configurations ranging from medium-sized bubbles up to the top of the barrier. Through this chain, a bubble can form in a small window, then grow as the window is enlarged, eventually becoming a stable bubble that does not easily disappear, thus avoiding directly crossing the huge full-volume condensation barrier.

Translation-invariant differentiable weighted order parameter

The original method employs a weighted order parameter on a fixed subvolume

OiW=1ViW∫ViWf(ϕ) dV,(4)\mathcal{O}_{i}^{W} = \frac{1}{V_{i}^{W}} \int_{V_{i}^{W}} f(\phi)\,\mathrm{d}V, \tag*{(4)}

where ViWV_{i}^{W} is a pre-selected spatial region [74]. However, the fixed window breaks translational invariance: the bubble tends to be “pinned” at the window center, so an additional shift update must be introduced to restore ergodicity of the bubble position. Moreover, the bubble may become stuck at the window edge during simulations, which could cause it to exceed the critical size and lead to simulation failure. Another approach is to locate the bubble by maximizing, over all possible centers, the convolution of the order parameter field with a weighting function, thereby making the weighted order parameter translation invariant without the shift update. This convolution can be computed efficiently via Fourier transform [74], but the max operation is generally nondifferentiable and cannot be used directly in HMC. To this end, we construct a translation-invariant, differentiable weighted order parameter Oinv\mathcal{O}_{\mathrm{inv}} that fully internalizes the translational degrees of freedom while retaining the differentiability required by HMC.

First, the field is smoothed once; the smoothed field ϕs\phi_{s} preserves spatial features such as the bubble radius and interface position, while suppressing high-frequency fluctuations in the metastable phase. To mark the stable-phase region, we introduce the weight

W(ϕs)=12[1+tanh⁡(α(ϕs−ϕ0W))],ϕ0W=ϕmn+ϕmx2,(5)W(\phi_{s}) = \frac{1}{2}\left[1 + \tanh\left(\alpha(\phi_{s} - \phi_{0}^{W})\right)\right], \qquad\phi_{0}^{W} = \frac{\phi_{mn} + \phi_{mx}}{2}, \tag*{(5)}

where OmnO_{mn} and OmxO_{mx} are the peak positions of the metastable and stable phases, respectively, and α\alpha controls the sharpness of the transition of the weight from the metastable to the stable phase. We take α=10\alpha= 10, which maintains the distinction between the two phases while avoiding a transition that is too steep and would increase the numerical stiffness of the HMC force. The zero point ϕ0W\phi_{0}^{W} is taken at the midpoint of the two phases, so that W(Omn)≈0W(O_{mn}) \approx0 and W(Omx)≈1W(O_{mx}) \approx1, thereby strongly suppressing the metastable phase in the subsequent centroid encoding. For each direction k=x,y,zk = x, y, z, we define the complex amplitude

Zk=∑xw(ϕs,x)ei2πxk/L=Xk+iYk.(6)Z_{k} = \sum_{x} w(\phi_{s,x})e^{i2\pi x_{k}/L} = X_{k} + iY_{k}. \tag*{(6)}

For a single-bubble configuration, the phase of ZkZ_{k} directly encodes the bubble centroid position, and the modulus ∣Zk∣\lvert Z_{k}\rvert reflects the bubble volume. To avoid numerical singularities when ∣Zk∣→0\lvert Z_{k}\rvert\to0 (no bubble), we introduce the regularization

Mk=Xk2+Yk2+ϵ2,uk=XkMk,vk=YkMk,(7)M_{k} = \sqrt{X_{k}^{2} + Y_{k}^{2} + \epsilon^{2}}, \qquad u_{k} = \frac{X_{k}}{M_{k}}, \qquad v_{k} = \frac{Y_{k}}{M_{k}}, \tag*{(7)}

where ϵ2=1\epsilon^{2} = 1 is a regularization constant with dimensions of lattice units. The resulting direction vector (uk,vk)(u_{k}, v_{k}) is a continuous, smooth encoding of the bubble centroid on the periodic torus. Broadcasting the direction vector to each lattice site yields the position field

h(x)=∑k[ukcos⁡2πxkL+vksin⁡2πxkL](8)h(x)=\sum_k\left[u_k\cos\frac{2\pi x_k}{L}+v_k\sin\frac{2\pi x_k}{L}\right] \tag*{(8)}
=∑k∣Zk∣Mkcos⁡(2πxkL−θk),(9)=\sum_k\frac{|Z_k|}{M_k}\cos\left(\frac{2\pi x_k}{L}-\theta_k\right), \tag*{(9)}

where θk=arg⁡Zk\theta_k=\arg Z_k. h(x)h(x) represents the proximity of lattice site xx to the bubble center. When the whole configuration is translated, h(x)h(x) shifts accordingly while maintaining its shape. In the ideal single-bubble case, it approximately satisfies

h(x)≃3−2π2L2∑k(xk−xc,k)2,(10)h(x)\simeq3-\frac{2\pi^2}{L^2}\sum_k(x_k-x_{c,k})^2, \tag*{(10)}

where xcx_c denotes the centroid position and h(x)h(x) reaches a maximum close to 3 at the centroid and decreases monotonically outward. Then, based on h(x)h(x), we construct the soft window

g(x)=12[1+tanh⁡(β(h(x)−c))],(11)g(x)=\frac{1}{2}\left[1+\tanh\left(\beta(h(x)-c)\right)\right], \tag*{(11)}

where

c=3.0−2π2σwindow2L2,β=L28π2σwindow.(12)c=3.0-\frac{2\pi^2\sigma_{\mathrm{window}}^2}{L^2},\qquad \beta=\frac{L^2}{8\pi^2\sigma_{\mathrm{window}}}. \tag*{(12)}

σwindow\sigma_{\mathrm{window}} is the chosen subensemble window radius. Thus, at r=σwindowr=\sigma_{\mathrm{window}}, g=1/2g=1/2, so the metastable background outside the subvolume is excluded from the order parameter calculation.

Finally, the translation-invariant differentiable weighted order parameter is defined as

Oinv(ϕ)=∑xg(x;ϕs)f(ϕs,x)+ϵreg1V∑xf(ϕs,x)∑xg(x;ϕs)+ϵreg(13)\mathcal{O}_{\mathrm{inv}}(\phi)= \frac{\sum_x g(x;\phi_s)f(\phi_s,x)+\epsilon_{\mathrm{reg}}\frac{1}{V}\sum_x f(\phi_s,x)} {\sum_x g(x;\phi_s)+\epsilon_{\mathrm{reg}}} \tag*{(13)}

where V=L3V=L^3, f(ϕs)=(ϕs−ϕ0f)∣ϕs−ϕ0f∣f(\phi_s)=(\phi_s-\phi_0^f)|\phi_s-\phi_0^f| is the local contribution function, and ϕ0f=Omn\phi_0^f=O_{mn} makes the zero point of the order parameter correspond to the pure metastable background. The regularization term ϵreg=10−3\epsilon_{\mathrm{reg}}=10^{-3} is a dimensionless numerical constant, used to prevent the denominator from becoming too small when the window is completely closed; when the window is completely closed, Oinv\mathcal{O}_{\mathrm{inv}} reduces to the global average 1V∑xf(ϕs,x)\frac{1}{V}\sum_x f(\phi_s,x).

Under periodic boundary conditions, the window center automatically follows the bubble, so Oinv\mathcal{O}_{\mathrm{inv}} remains unchanged when the bubble drifts freely in the system. It can also be verified that Oinv∈C1(ϕ)\mathcal{O}_{\mathrm{inv}}\in C^1(\phi), that its gradient ∇ϕOinv\nabla_\phi\mathcal{O}_{\mathrm{inv}} is continuous, and that it is accurately computed via adjoint propagation. The forward smoothing and the adjoint kernel strictly match, the forces can be obtained analytically, and HMC detailed balance is strictly preserved. Therefore, this order parameter is naturally suited for multicanonical HMC (MCHMC) updates in the expanded ensemble method. It avoids both the fixed-window limitation of the original method and the non-differentiability of the Fourier transform maximum scheme, providing a simpler and more efficient route for expanded ensembles within the HMC framework.

Update algorithms

To move the system along the chain, the algorithm combines two types of updates. Within each subensemble ii, standard multicanonical updates are used to change the field configuration and explore the allowed order-parameter range; the weight WiW_i is designed to flatten the histogram of the weighted order parameter OiW\mathcal{O}_i^W, which denotes Oinv\mathcal{O}_{\mathrm{inv}} within subensemble ii. In addition, ensemble swap updates attempt to swap between adjacent subensembles ii and jj. The acceptance probability for such a swap is

P(i→j)=min⁡(1,exp⁡[−(Wj(OjW)−Wi(OiW)−(ηj−ηi))]),(14)P(i\to j)=\min\left(1,\exp\left[-\left(W_j(\mathcal{O}_j^W)-W_i(\mathcal{O}_i^W)-(\eta_j-\eta_i)\right)\right]\right), \tag*{(14)}

where ηi\eta_i are bias parameters adjusted in a later iteration to balance the visitation probabilities of the different subensembles. When the configuration lies at the boundary of the overlap region of the two ensembles, the acceptance probability is close to unity. A schematic illustration of the ensemble swap update is shown in Figure 2.

Illustration of the ensemble swap update

Figure 2. Illustration of the ensemble swap update between adjacent subensembles.

Iteration procedure

The iteration employs a two-step iterative procedure. First, the multicanonical weights WiW_i are iterated independently for each subensemble ii to obtain the respective weight functions to ensure each subensemble uniformly visits its allowed order-parameter range. After fixing all WiW_i, the relative weights ηi\eta_i are iteratively adjusted so that the system visits each subensemble index ii with equal probability, ensuring that all subensembles on the chain are visited equally, thereby obtaining global statistics that are as accurate as possible.

Recovering the physical probability

After the above procedure, one obtains the multicanonical probability distribution PmucP_{\mathrm{muc}}. By reweighting, the canonical (physical) probability distribution PcanP_{\mathrm{can}} is obtained. In the simulation, the condensation barrier only appears in the first small-volume subensemble, where it is negligible. Therefore, the total simulation time does not grow exponentially with the volume, and the statistical part of the nucleation rate is

Pcnormalised=P(∣O−Oc∣<ϵ/2)ϵP(O<Oc).(15)P_c^{\mathrm{normalised}}=\frac{P(\lvert\mathcal{O}-\mathcal{O}_c\rvert<\epsilon/2)}{\epsilon P(\mathcal{O}<\mathcal{O}_c)}. \tag*{(15)}

Real-time Langevin dynamics and complete nucleation rate

The statistical weight gives the exponential suppression factor, but to obtain the complete nucleation rate, one must compute the dynamical factor near the critical point [38, 71]. Based on the critical field configurations collected from the multicanonical simulation, the real-time dynamics is governed by the Langevin equations in Eq. (41), with the effective Hamiltonian

Heff=∫d3x[12π2+12(∂iϕ)2+V3(ϕ)].(16)H_{\mathrm{eff}}=\int d^{3}x\left[\frac{1}{2}\pi^{2}+\frac{1}{2}(\partial_i\phi)^2+V_3(\phi)\right]. \tag*{(16)}

For each critical configuration, we compute the flux across the barrier,

⟨flux⟩=⟨∣ΔOΔt∣⟩Oc,(17)\langle\mathrm{flux}\rangle=\left\langle\left|\frac{\Delta\mathcal{O}}{\Delta t}\right|\right\rangle_{\mathcal{O}_c}, \tag*{(17)}

which represents the average rate of change of the order parameter near the critical value Oc\mathcal{O}_c. We employ the smoothed nonlinear order parameter

O[ϕ]=1V∑x(ϕs(x)−A)∣ϕs(x)−A∣,ϕs=Sϕ,(18)\mathcal{O}[\phi]=\frac{1}{V}\sum_x(\phi_s(x)-A)\lvert\phi_s(x)-A\rvert,\qquad\phi_s=S\phi, \tag*{(18)}

where SS is the separable binomial smoothing operator and AA is the peak position of the metastable-phase histogram. Since the momentum field is Gaussian, ⟨π(x)π(y)⟩=δxy\langle\pi(x)\pi(y)\rangle=\delta_{xy}, the conditional distribution of O˙\dot{\mathcal{O}} at fixed ϕ\phi is Gaussian with zero mean and variance

σ2=1V2∑x[Sf′(ϕs−A)]x2,(19)\sigma^2=\frac{1}{V^2}\sum_x\left[S f'(\phi_s-A)\right]_x^2, \tag*{(19)}

where f(u)=u∣u∣f(u)=u\lvert u\rvert, so that f′(u)=2∣u∣f'(u)=2\lvert u\rvert. Here the subscript xx denotes the value of the field obtained after applying SS. The flux through the critical surface O=Oc\mathcal{O}=\mathcal{O}_c is therefore

⟨∣ΔOΔt∣⟩Oc=8πV2⟨∑x[S(∣ϕs−A∣)]x2⟩Oc.(20)\left\langle\left|\frac{\Delta\mathcal{O}}{\Delta t}\right|\right\rangle_{\mathcal{O}_c}=\sqrt{\frac{8}{\pi V^2}}\left\langle\sqrt{\sum_x\left[S(\lvert\phi_s-A\rvert)\right]_x^2}\right\rangle_{\mathcal{O}_c}. \tag*{(20)}

where ⟨⋯ ⟩Oc\langle\cdots\rangle_{\mathcal{O}_c} denotes the average over configurations constrained to O=Oc\mathcal{O}=\mathcal{O}_c. In the unsmoothed case S=IS=I, Eq. (20) reduces to

⟨∣ΔOΔt∣⟩Oc=8πV⟨1V∑x(ϕ(x)−A)2⟩Oc.(21)\left\langle\left|\frac{\Delta\mathcal{O}}{\Delta t}\right|\right\rangle_{\mathcal{O}_c}=\sqrt{\frac{8}{\pi V}}\left\langle\sqrt{\frac{1}{V}\sum_x(\phi(x)-A)^2}\right\rangle_{\mathcal{O}_c}. \tag*{(21)}

For the quadratic order parameter ϕav2\phi_{\mathrm{av}}^2, taking A=0A=0 and positive fields, one recovers the familiar result [78]

⟨∣ϕ˙av2∣⟩ϕav,c2=8ϕav,c2πV.(22)\left\langle\left|\dot{\phi}_{\mathrm{av}}^2\right|\right\rangle_{\phi_{\mathrm{av},c}^2}=\sqrt{\frac{8\phi_{\mathrm{av},c}^2}{\pi V}}. \tag*{(22)}

In addition, each configuration is evolved both forward and backward in time, with the initial momenta reversed for the backward evolution. We count the fraction of trajectories that truly cross the barrier by defining

d=δtunnelNcrossings,(23)d=\frac{\delta_{\mathrm{tunnel}}}{N_{\mathrm{crossings}}}, \tag*{(23)}

where δtunnel=1\delta_{\mathrm{tunnel}}=1 if the trajectory starts and ends in different phases and 0 otherwise, and NcrossingsN_{\mathrm{crossings}} is the number of times the trajectory crosses Oc\mathcal{O}_c. A factor of 1/21/2 is included to count only metastable →\to stable trajectories, avoiding double counting. Although ⟨∣ΔO/Δt∣⟩\langle\lvert\Delta\mathcal{O}/\Delta t\rvert\rangle and dd individually depend on the Langevin time step, their product converges in the short-time limit.

From the Langevin simulations of critical configurations, we measure the velocity factor ⟨flux⟩\langle\mathrm{flux}\rangle and the transmission coefficient dd. The product of these two quantities gives the dynamical prefactor. Although the individual factors vary when the step is halved, their product remains constant within errors, as expected from the factorization of the nucleation rate in Langer’s theory [38]. The dynamical factor AdynA_{\mathrm{dyn}} is decomposed into two physically distinct quantities:

Adyn↔⟨flux⟩×d.(24)A_{\mathrm{dyn}}\leftrightarrow\langle\mathrm{flux}\rangle\times d. \tag*{(24)}

Therefore, the final form of the nucleation rate in the numerical simulation is [78, 79]:

Γ=12VPcnormalised⟨∣ΔOΔt∣⟩d.(25)\Gamma=\frac{1}{2V}P_c^{\mathrm{normalised}}\left\langle\left|\frac{\Delta\mathcal{O}}{\Delta t}\right|\right\rangle d. \tag*{(25)}

The flux-over-population construction ensures that the final rate is independent of the order parameter, as long as the dividing surface separates the two phases [71]. We therefore regard PcnormalisedP_c^{\mathrm{normalised}}, ⟨∣ΔO/Δt∣⟩\langle\lvert\Delta\mathcal{O}/\Delta t\rvert\rangle, and dd as coordinate-dependent intermediate quantities and do not attach separate physical meaning to them.

Lattice details

In this section, we present the model for lattice simulation and the simulation setup.

Lattice action and parameters

To investigate the bubble nucleation rate for the first-order phase transition in this work, the lattice 3D action takes the form

S3[ϕ]=∫d3x[12(∂iϕ)2+V3(ϕ)],(26)S_3[\phi]=\int d^3x\left[\frac{1}{2}(\partial_i\phi)^2+V_3(\phi)\right], \tag*{(26)}

with the 3D effective potential [80]:

V3(ϕ)=12m32ϕ2+16g3ϕ3+124λ3ϕ4,(27)V_{3}(\phi) = \frac{1}{2}m_{3}^{2}\phi^{2} + \frac{1}{6}g_{3}\phi^{3} + \frac{1}{24}\lambda_{3}\phi^{4}, \tag*{(27)}

where

m32=γ(T32−(T30)2),g3=−2αT33/2,λ3=6λT3.(28)m_{3}^{2} = \gamma(T_{3}^{2} - (T_{3}^{0})^{2}), \quad g_{3} = -2\alpha T_{3}^{3/2}, \quad\lambda_{3} = 6\lambda T_{3}. \tag*{(28)}

The 3D dimensionless parameters used in this work are

γ=1.17,αT3=1.45,λT31/2=0.5,\gamma= 1.17,\quad\alpha T_{3} = 1.45,\quad\lambda T_{3}^{1/2} = 0.5,
T30=0.80,T3=1.49.(29)T_{3}^{0} = 0.80,\quad T_{3} = 1.49. \tag*{(29)}

Figure 3 compares the potential used in this work with that of Ref. [79]. Although the two potentials are not normalised to the same dimensionless scale, the barrier obtained here is higher, indicating that nucleation is more strongly suppressed.

Figure 3. This shows the difference between the potential used in this work and that in Ref. [79]. The blue, green, and red dots indicate the false vacuum, the barrier, and the true vacuum, respectively.

Considering the influence of lattice discretisation, the lattice action takes the form

Slat=∑x[−12Zϕϕx(∇lat2ϕ)x+σlatϕx+12ZϕZmmlat2ϕx2+14!Zϕ2λlatϕx4].(30)S_{\mathrm{lat}} = \sum_{x}\left[-\frac{1}{2}Z_{\phi}\phi_{x}(\nabla_{\mathrm{lat}}^{2}\phi)_{x} + \sigma_{\mathrm{lat}}\phi_{x} + \frac{1}{2}Z_{\phi}Z_{m}m_{\mathrm{lat}}^{2}\phi_{x}^{2} + \frac{1}{4!}Z_{\phi}^{2}\lambda_{\mathrm{lat}}\phi_{x}^{4}\right]. \tag*{(30)}

where ZϕZ_{\phi} and ZmZ_{m} are multiplicative renormalisation constants, and σlat\sigma_{\mathrm{lat}} is the coefficient of the linear term after the shift. The counterterms δλ\delta\lambda and δm2\delta m^{2} are included in the definitions of λlat\lambda_{\mathrm{lat}} and mlat2m_{\mathrm{lat}}^{2}. The derivation of the dimensionally reduced action, the field shift, the matching relations, and the overcooling procedure is presented in Appendix D.

Simulation setup

Here we show the lattice geometry, the spacing choices, and the sampling parameters of both the expanded-ensemble MCHMC simulation and the real-time Langevin simulation.

Lattice geometry and spacing

All simulations use periodic boundary conditions on a cubic lattice with Ns3N_{s}^{3} sites and physical volume L3L^{3}. We keep the dimensionless physical volume approximately fixed, Lλ3∼128L\lambda_{3}\sim128, so that the lattice spacing aλ3a\lambda_{3} and the number of sites are related as

Ns=La=Lλ3aλ3.(31)N_{s} = \frac{L}{a} = \frac{L\lambda_{3}}{a\lambda_{3}}. \tag*{(31)}

The spacings used in this work are aλ3∈{1.5,2.0,2.5,3.0}a\lambda_{3} \in\{1.5, 2.0, 2.5, 3.0\}, which correspond to Ns=86,64,52,42N_{s} = 86, 64, 52, 42 respectively. The range of aλ3a\lambda_{3} is chosen so that the O(a3)O(a^{3}) continuum extrapolation is reliable. The smallest spacing is limited by the cost of increasing Ns=L/aN_{s}=L/a and also by the validity of the high-temperature dimensional reduction, which requires

a≳12πT∼1T.(32)a \gtrsim\frac{1}{2\pi T} \sim\frac{1}{T}. \tag*{(32)}

For our parameters T3=1.49T_{3}=1.49, this bound corresponds to aλ3≳1.0a\lambda_{3}\gtrsim1.0.

aλ3a\lambda_{3}rini/ar_{\mathrm{ini}}/aα\alphaNensN_{\mathrm{ens}}Δt\Delta tNMDN_{\mathrm{MD}}
1.561.1590.1215
2.061.1570.1210
2.551.1530.108
3.051.1060.105

Table 1. Statistical lattice simulation parameters for each spacing.

Expanded-ensemble MCHMC parameters

The expanded-ensemble chain consists of NensN_{\mathrm{ens}} subensembles, while subensembles vary with the lattice volume. The first subensemble uses the smallest subvolume, taken to be a sphere of radius rinir_{\mathrm{ini}}, and subsequent subensembles enlarge the subvolume by a fixed growth ratio α=ri+1/ri\alpha=r_{i+1}/r_{i}, so that the ii-th subensemble uses a sphere of radius ri=rini(α)i−1r_{i}=r_{\mathrm{ini}}(\alpha)^{i-1}, increasing until the subcritical size is reached. The molecular-dynamics parameters also depend on the lattice spacing and are tuned to balance simulation efficiency against maintaining an average HMC acceptance rate of about 60%. Here Δt\Delta t is the HMC molecular-dynamics time step and NMDN_{\mathrm{MD}} is the number of molecular-dynamics steps per trajectory. The corresponding per-spacing parameters are collected in Table I. The production runs contain around Nconf∼106N_{\mathrm{conf}}\sim10^{6} configurations per subensemble. Statistical errors are estimated by jackknife resampling over contiguous time blocks of the expanded-ensemble chain: the chain is divided into 20 blocks, and each jackknife sample removes the same time block from all subensembles before recomputing PcnormalisedP_c^{\mathrm{normalised}}.

configurations.

Real-time Langevin parameters

The dynamical prefactor is extracted from real-time Langevin evolution of critical configurations. For each lattice spacing, we use Ncrit∼150N_{\mathrm{crit}} \sim150 critical configurations, which selected from the multicanonical simulation in the interval ∣O−Oc∣<ϵ/2|\mathcal{O} - \mathcal{O}_c| < \epsilon/2. The Langevin equations are integrated with time step ΔtLang=0.05\Delta t_{\mathrm{Lang}} = 0.05, using a damping coefficient γ=1/L\gamma= 1/L [78]. For each configuration, we perform 30 independent Langevin runs, each evolved forward and then backward, with the momenta reversed at the start of the backward evolution. The statistical error on the ⟨flux⟩\langle\mathrm{flux}\rangle is estimated from the config-to-config scatter over the NcritN_{\mathrm{crit}} configurations, while that on dd is estimated from a Poisson bootstrap over the YES/NO outcomes.

Results

In this section we present the numerical results for the statistical factor, the dynamical factor, and the final nucleation rate. The statistical factor is obtained from the expanded-ensemble MCHMC simulation and extrapolated to the continuum limit, while the dynamical factor is extracted from real-time Langevin evolution of critical configurations.

Statistical factor

The reweighted physical probability distribution P(O)P(\mathcal{O}) at aλ3=3a\lambda_3 = 3 is shown in Figure 4. The different curves correspond to the individual subensembles of the expanded-ensemble chain, which overlap in the order-parameter range and are combined into a single continuous distribution after normalisation. The left peak corresponds to the symmetric phase, while the intermediate region around the local minimum of PP is the droplet branch from which the critical bubble configuration is extracted. The normalised critical-bubble probability PcnormalisedP_c^{\mathrm{normalised}} is obtained from Eq. (15).

Reweighted physical probability distribution $P(\mathcal{O})$

Figure 4. Reweighted physical probability distribution P(O)P(\mathcal{O}) obtained from the expanded-ensemble multicanonical simulation with aλ3=3a\lambda_3 = 3. The different curves correspond to individual subensembles, which are combined after normalisation.

Since the lattice action is O(a4)O(a^4)-improved, the leading lattice-spacing dependence of the logarithm of the rate arises at O(a3)O(a^3). A least-squares fit of the form

log⁡10Pcnormalised=f0+c(aλ3)3(33)\log_{10} P_c^{\mathrm{normalised}} = f_0 + c(a\lambda_3)^3 \tag*{(33)}

to the four data points yields, in the continuum limit,

log⁡10Pcnormalised=−150.039±0.025,(34)\log_{10} P_c^{\mathrm{normalised}} = -150.039 \pm0.025, \tag*{(34)}

where the error has been inflated by χ2/dof\sqrt{\chi^2/\mathrm{dof}} to account for the fit quality. The best-fit line and the lattice data are shown in Figure 5.

Zero-lattice-spacing extrapolation with cubic fits

Figure 5. The zero-lattice-spacing extrapolation with cubic fits.

Dynamical factor

The dynamical factor AdynA_{\mathrm{dyn}} is extracted from real-time Langevin evolution of critical configurations, as described in Sec. II D. For each lattice spacing, we select Ncrit∼150N_{\mathrm{crit}}\sim150 configurations in the interval ∣O−OC∣<ϵ/2\left|O-O_C\right|<\epsilon/2 and evolve each configuration both forward and backward in time, with initial momenta reversed for the backward evolution. From these trajectories we determine the flux ⟨∣ΔO/Δt∣⟩\langle\left|\Delta O/\Delta t\right|\rangle and the effective tunnelling fraction dd.

Figure 6 shows the resulting dynamical factor at different lattice spacings. The flux depends smoothly on the lattice spacing; within errors, it is constant over the range. The tunnelling fraction dd, on the other hand, decreases as the lattice spacing is reduced. At the smallest spacing, aλ3=1.5a\lambda_3=1.5, we evolved 800 critical configurations, producing 24 000 real-time trajectories in total, none of which successfully completes a full metastable-to-stable crossing; instead, they eventually run into the true vacuum. We conjecture the reason is that dd at aλ3=1.5a\lambda_3=1.5 is so low that it cannot be detected in this simulation. The factor dd has been take to be a constant in Ref. [78].

Flux, tunnelling fraction, and dynamical factors at different lattice spacings

Figure 6. ⟨flux⟩\langle\mathrm{flux}\rangle, dd and dynamical factors at different lattice spacings. The upper panel shows the ⟨flux⟩\langle\mathrm{flux}\rangle, the middle panel shows dd, and the lower panel shows the full dynamical factor.

The continuum-extrapolated nucleation rate per unit volume is

Γ=12VPcnormalised⟨∣ΔO/Δt∣⟩d=1.41×10−158.(35)\Gamma=\frac{1}{2V}P_c^{\mathrm{normalised}}\left\langle\left|\Delta O/\Delta t\right|\right\rangle d=1.41\times10^{-158}. \tag*{(35)}

Final nucleation rate

Combining the statistical and dynamical factors yields the final nucleation rate, shown in Figure 7. Because of the limited dynamical factor data, we can fit only three points, and the fit quality is poor; nevertheless, it can still be seen that the nucleation rate decreases as aλ3a\lambda_3 decreases, which is opposite to the trend in Ref. [79].

Zero-lattice-spacing extrapolation of the final nucleation rate

Figure 7. The zero-lattice-spacing extrapolation of the final nucleation rate.

Using the numerical package BubbleDet [81], we compute the full one-loop theoretical result:

e−ΔS[ϕS]≃9.99×10−113,Apre≃2.00×10−33,(36)e^{-\Delta S[\phi_S]}\simeq9.99\times10^{-113},\quad A_{\mathrm{pre}}\simeq2.00\times10^{-33}, \tag*{(36)}
Astat=Apree−ΔS[ϕS]≃2.00×10−145,(37)A_{\mathrm{stat}}=A_{\mathrm{pre}}e^{-\Delta S[\phi_S]}\simeq2.00\times10^{-145}, \tag*{(37)}
Adyn≃4.52×10−2.(38)A_{\mathrm{dyn}}\simeq4.52\times10^{-2}. \tag*{(38)}

The corresponding theoretical nucleation rate is

Γ1−loop=Astat×Adyn=9.04×10−147.(39)\Gamma^{\mathrm{1-loop}}=A_{\mathrm{stat}}\times A_{\mathrm{dyn}}=9.04\times10^{-147}. \tag*{(39)}

See Appendix A for details. Hence the difference in ∣log⁡Γ∣\left|\log\Gamma\right| is

∣log⁡ΓNP−log⁡Γ1−loop∣∣log⁡Γ1−loop∣=8.09%,(40)\frac{\left|\log\Gamma^{\mathrm{NP}}-\log\Gamma^{\mathrm{1-loop}}\right|}{\left|\log\Gamma^{\mathrm{1-loop}}\right|}=8.09\%, \tag*{(40)}

This is consistent with the finding of Ref. [79] that one-loop perturbation theory and lattice simulations differ by about 20% in ∣log⁡Γ∣\left|\log\Gamma\right| in a real scalar theory with a tree-level barrier, even in a parameter region where perturbation theory is expected to perform well. Part of this difference may be due to the one-loop approximation, and a detailed comparison is left for future work.

Conclusion and Discussion

We present a fully nonperturbative calculation of the bubble nucleation rate for a first-order phase transition in a three-dimensional scalar field theory with an asymmetric double-well potential. In doing so, we extend the expanded-ensemble multicanonical method to three dimensions for the first time. The statistical prefactor is obtained from HMC updates [75], while the dynamical prefactor is extracted from real-time Langevin evolution. Combining the two yields the full nucleation rate, which we compare with the perturbative result computed with BubbleDet [81].

Although we retain the standard decomposition of the nucleation rate into a statistical and a dynamical factor [71, 78, 82], our method overcomes the condensation barrier and the sampling bottleneck that plague standard multicanonical simulations at large volumes by employing an expanded-ensemble MCHMC with a translation-invariant, differentiable weighted order parameter. Within the same three-dimensional real scalar framework as in Ref. [79], we additionally consider a different potential and a different choice of parameters, thereby providing a nonperturbative benchmark that is independent of the perturbative calculation and allowing the statistical and dynamical contributions to be separated systematically. The resulting nonperturbative rate, Γ/V≃1.41×10−158\Gamma/V \simeq1.41 \times10^{-158} in lattice units, differs from the BubbleDet result by about 8% in ∣log⁡Γ∣|\log\Gamma|. This is of the same order as the difference of about 20% in ∣log⁡Γ∣|\log\Gamma| reported in Ref. [79] for the parameter choices considered there. Such comparisons are useful for testing the convergence of perturbation theory and provide a well-defined target for future two-loop calculations.

Physically, the nucleation rate obtained in this work provides a nonperturbative input for predictions of the gravitational-wave signal from first-order electroweak phase transitions, helping to reduce theoretical uncertainties relevant to LISA and pulsar-timing-array observations [25, 26, 27, 28, 29, 30]. It also governs the degree of supercooling and the duration of the transition, and therefore influences the efficiency of electroweak baryogenesis [3, 4, 5].

Methodologically, the expanded-ensemble HMC framework can in principle be extended to other rare processes characterized by a condensation barrier and a suitable differentiable order parameter, such as topological-defect formation and nucleation in more complex gauge–Higgs systems.

The present study is restricted to scalar fields and does not include gauge fields or fermions. A more systematic investigation of the continuum limit and of finite-volume effects [77] is likewise desirable. Future applications of the method include its extension to more realistic electroweak models, comparison of the results with two-loop perturbative calculations, and the use of the nonperturbative nucleation rate as an input for gravitational-wave spectrum simulations, enabling further cosmological predictions and tests.

Acknowledgement

This work was supported in part by the National Natural Science Foundation of China under Grant Nos. 12675130, 12322505, 12547101, and Fundamental Research Funds for the Central Universities Grant No. 2026CDJZKPT-10. L.B. also acknowledges Chongqing Natural Science Foundation under Grant No. CSTB2024NSCQ-JQX0022. We also acknowledge the use of the HILA lattice framework for the numerical simulations presented in this work and thank its developers for making it publicly available.

Review of Langer theory

For a thermal first-order phase transition, the bubble nucleation rate can be described by Langer’s finite-temperature field theory [37, 38, 39, 40]. The long-wavelength dynamics of the scalar field is governed by the Langevin equations

∂tϕ=π,∂tπ=−δHeffδϕ−γπ+ξ,(41)\partial_t \phi= \pi,\qquad\partial_t \pi= -\frac{\delta H_{\mathrm{eff}}}{\delta\phi} - \gamma\pi+ \xi, \tag*{(41)}

where ϕ\phi is the 3D effective scalar field, π\pi its canonical momentum, HeffH_{\mathrm{eff}} the effective Hamiltonian constructed to match the long-wavelength equal-time correlation functions of the full quantum theory [80, 83], γ\gamma the damping rate, and ξ\xi a Gaussian noise satisfying the fluctuation-dissipation relation

⟨ξ(t,x)ξ(t′,x′)⟩=2γδ(t−t′)δ(3)(x−x′).(42)\langle\xi(t,\boldsymbol{x}) \xi(t',\boldsymbol{x}') \rangle= 2\gamma\delta(t-t') \delta^{(3)}(\boldsymbol{x}-\boldsymbol{x}'). \tag*{(42)}

The probability density P(ϕ,π)P(\phi,\pi) obeys a Fokker-Planck equation [83]. Assuming the metastable phase is sufficiently long-lived, the nucleation rate per unit volume is given by the steady probability flux over the barrier [38, 84] and factorizes as

Γ=Adyn×Astat,(43)\Gamma= A_{\mathrm{dyn}} \times A_{\mathrm{stat}}, \tag*{(43)}

where AstatA_{\mathrm{stat}} is the statistical factor, corresponding to the determinant of fluctuations around the critical bubble, and AdynA_{\mathrm{dyn}} is the dynamical factor, relating to the flux at the top of the barrier and the transmission probability [80, 82, 83].

The statistical factor is obtained from a saddle-point expansion about the critical bubble ϕS\phi_S [35, 36, 41, 42],

Astat=Apree−ΔS[ϕS],(44)A_{\mathrm{stat}} = A_{\mathrm{pre}} e^{-\Delta S[\phi_S]}, \tag*{(44)}

where the fluctuation prefactor is

Apre=(ΔS[ϕS]2π)3/2∣det⁡(S′′[ϕ0]/2π)det⁡′(S′′[ϕS]/2π)∣,(45)A_{\mathrm{pre}} = \left(\frac{\Delta S[\phi_S]}{2\pi}\right)^{3/2} \sqrt{\left|\frac{\det(S''[\phi_0]/2\pi)}{\det'(S''[\phi_S]/2\pi)}\right|}, \tag*{(45)}

where ϕ0\phi_0 is the metastable phase, S′′S'' denotes the second functional derivative of the action, the prime on det⁡′\det' means that zero modes are omitted [81], and

ΔS[ϕS]≡S[ϕS]−S[ϕ0].(46)\Delta S[\phi_S] \equiv S[\phi_S] - S[\phi_0]. \tag*{(46)}

The dynamical factor is

Adyn=12π(∣λ−∣+γ24−γ2),(47)A_{\mathrm{dyn}} = \frac{1}{2\pi}\left(\sqrt{\lvert\lambda_{-}\rvert+\frac{\gamma^{2}}{4}}-\frac{\gamma}{2}\right), \tag*{(47)}

where λ−\lambda_{-} is the single negative eigenvalue of S′′[ϕS]S^{\prime\prime}[\phi_{S}] and γ\gamma is the damping rate [39, 80, 82]. These expressions are the one-loop perturbative results; in our lattice calculation we extract the nonperturbative analogues directly [71, 78, 82].

Free energy

The reweighted probability distribution P(O)P(\mathcal{O}) is shown in Figure 4. The free energy is defined as

F(O)=−Tln⁡P(O),(48)F(\mathcal{O})=-T\ln P(\mathcal{O}), \tag*{(48)}

where O\mathcal{O} is an order parameter that measures the degree of the phase transition and distinguishes the symmetric and broken phases. In the dimensionally reduced 3D effective theory, the partition function is

Z=∫Dϕe−S3[ϕ],(49)Z=\int\mathcal{D}\phi e^{-S_{3}[\phi]}, \tag*{(49)}

with S3S_{3} the dimensionless 3D effective action. The constrained free energy is then [69, 70]

F(O0)=−Tln⁡∫Dϕe−S3[ϕ]δ(O−O0).(50)F(\mathcal{O}_{0})=-T\ln\int\mathcal{D}\phi e^{-S_{3}[\phi]}\delta(\mathcal{O}-\mathcal{O}_{0}). \tag*{(50)}

The difference between the free energy at the separatrix and that of the metastable minimum defines the dimensionless suppression exponent

S≡FC−FsymmT,(51)S\equiv\frac{F_{C}-F_{\mathrm{symm}}}{T}, \tag*{(51)}

where FCF_{C} is the free energy at the separatrix, corresponding to the critical bubble configuration, and FsymmF_{\mathrm{symm}} is the free energy of the metastable phase.

In the saddle-point approximation, the free-energy difference is given by the classical action difference,

FC−Fsymm≈TΔS[ϕS].(52)F_{C}-F_{\mathrm{symm}}\approx T\Delta S[\phi_{S}]. \tag*{(52)}

Including one-loop fluctuations around the saddle point ϕS\phi_{S} [35, 36], this becomes

FC−Fsymm≃TΔS[ϕS]−Tln⁡[(ΔS[ϕS]2π)3/2×∣det⁡′[−∇2+V3′′(ϕS)]det⁡[−∇2+V3′′(ϕ0)]∣−1/2]+O(ℏ).(53)\begin{aligned} F_{C}-F_{\mathrm{symm}} &\simeq T\Delta S[\phi_{S}]-T\ln\left[\left(\frac{\Delta S[\phi_{S}]}{2\pi}\right)^{3/2}\right.\\ &\left.\times\left\lvert\frac{\det^{\prime}[-\nabla^{2}+V_{3}^{\prime\prime}(\phi_{S})]}{\det[-\nabla^{2}+V_{3}^{\prime\prime}(\phi_{0})]}\right\rvert^{-1/2}\right]+O(\hbar). \tag*{(53)} \end{aligned}

Here V3′′V_{3}^{\prime\prime} is the second derivative of the 3D effective potential. Thus, the probability ratio contains both the classical action and the one-loop fluctuation correction, without a separate determinant calculation. Numerically, the statistical factor AstatA_{\mathrm{stat}} is extracted from the difference between the measured suppression exponent SS and the classical action difference ΔS[ϕS]\Delta S[\phi_{S}] [71, 78].

Review of multicanonical sampling

Multicanonical Monte Carlo Method

Let the Hamiltonian of the system be HH, and choose a suitable order parameter O\mathcal{O}. In the canonical ensemble, the probability density of the order parameter O\mathcal{O} is

Pcan(O)∝∫Dϕe−H[ϕ]δ(O[ϕ]−O).P_{\mathrm{can}}(\mathcal{O})\propto\int\mathcal{D}\phi e^{-H[\phi]}\delta(\mathcal{O}[\phi]-\mathcal{O}).

In phase transition or barrier regions, it is often exponentially suppressed, and direct sampling can hardly cover these regions. The multicanonical Monte Carlo method introduces a weight function W(O)W(\mathcal{O}) that depends only on the order parameter, equivalently defining the multicanonical potential,

V(O)=−ln⁡W(O)+const.V(\mathcal{O})=-\ln W(\mathcal{O})+\mathrm{const}.

The sampling weight is modified to

p∝e−HW(O)=e−H−V(O),p\propto e^{-H}W(\mathcal{O})=e^{-H-V(\mathcal{O})},

so the marginal distribution of the order parameter becomes

Pmuc(O)∝Pcan(O)e−V(O).P_{\mathrm{muc}}(\mathcal{O})\propto P_{\mathrm{can}}(\mathcal{O})e^{-V(\mathcal{O})}.

By iteratively adjusting the weight or the multicanonical potential, one ensures that within the target interval Pmuc(O)≈constP_{\mathrm{muc}}(\mathcal{O})\approx\mathrm{const}. The optimal form is

V(O)=ln⁡Pcan(O)+const,V(\mathcal{O})=\ln P_{\mathrm{can}}(\mathcal{O})+\mathrm{const},

or equivalently

W(O)∝1Pcan(O).W(\mathcal{O})\propto\frac{1}{P_{\mathrm{can}}(\mathcal{O})}.

In this way, the system is forced to perform an approximately uniform random walk within the required range of O\mathcal{O}, thereby obtaining statistical samples in the exponentially suppressed region. Finally, by reweighting the multicanonical samples with weight eV(O)e^{V(\mathcal{O})}, one recovers the physical probability distribution of the canonical ensemble:

Pcan(O)∝Pmuc(O)eV(O).P_{\mathrm{can}}(\mathcal{O})\propto P_{\mathrm{muc}}(\mathcal{O})e^{V(\mathcal{O})}.

Effective Action and Multicanonical Potential

Let the lattice field be ϕ\phi, the canonical action be S[ϕ]S[\phi], and the coupling constant be β\beta. Choose a globally differentiable order parameter O[ϕ]\mathcal{O}[\phi]. The goal is to produce a flat marginal distribution of the order parameter on the interval [Omin⁡,Omax⁡][\mathcal{O}_{\min},\mathcal{O}_{\max}].

In the canonical ensemble, the probability density of the order parameter O\mathcal{O} is

Pcan(O)∝∫Dϕe−βS[ϕ]δ(O[ϕ]−O).P_{\mathrm{can}}(\mathcal{O})\propto\int\mathcal{D}\phi e^{-\beta S[\phi]}\delta(\mathcal{O}[\phi]-\mathcal{O}).

Introducing the multicanonical potential V(O)V(\mathcal{O}), the effective action is taken as

S^[ϕ]=βS[ϕ]+V(O[ϕ]).\hat{S}[\phi] = \beta S[\phi] + V(\mathcal{O}[\phi]).

Under this effective action, the sampling weight is e−S^[ϕ]e^{-\hat{S}[\phi]}, and the corresponding marginal distribution of the order parameter is

PMUCA(O)∝Pcan(O)e−V(O).P_{\mathrm{MUCA}}(\mathcal{O}) \propto P_{\mathrm{can}}(\mathcal{O})e^{-V(\mathcal{O})}.

If it is required to be approximately constant in [Omin⁡,Omax⁡][\mathcal{O}_{\min},\mathcal{O}_{\max}], then the optimal multicanonical potential satisfies

V(O)=ln⁡Pcan(O)+const.V(\mathcal{O}) = \ln P_{\mathrm{can}}(\mathcal{O}) + \mathrm{const}.

In numerical implementations, V(O)V(\mathcal{O}) is unknown and must be determined through an iterative bootstrap process. Its parameterization can be chosen as piecewise linear or polynomial interpolation, written as

V(O)=β^(O)O+α^(O),V(\mathcal{O}) = \hat{\beta}(\mathcal{O})\mathcal{O} + \hat{\alpha}(\mathcal{O}),

where β^(O)\hat{\beta}(\mathcal{O}) and α^(O)\hat{\alpha}(\mathcal{O}) are undetermined piecewise functions.

Weight Iteration and Reweighting

The specific steps of weight iteration are as follows:

(a) Initial guess: First perform a short canonical HMC simulation to obtain the order-parameter histogram H0(O)H_0(\mathcal{O}). Let the initial weight be W(0)(O)∝1/H0(O)W^{(0)}(\mathcal{O}) \propto1/H_0(\mathcal{O}); the corresponding multicanonical potential is

V(0)(O)=−ln⁡W(0)(O)+const=ln⁡H0(O)+const.V^{(0)}(\mathcal{O}) = -\ln W^{(0)}(\mathcal{O}) + \mathrm{const} = \ln H_0(\mathcal{O}) + \mathrm{const}.

(b) Iterative update: Run MCHMC under the current V(i)(O)V^{(i)}(\mathcal{O}) to obtain a new order-parameter histogram H(i+1)(O)H^{(i+1)}(\mathcal{O}). Update the weight

W(i+1)(O)∝W(i)(O)×H(i+1)(O),W^{(i+1)}(\mathcal{O}) \propto W^{(i)}(\mathcal{O}) \times H^{(i+1)}(\mathcal{O}),

or equivalently update the multicanonical potential

V(i+1)(O)=V(i)(O)−ln⁡H(i+1)(O).V^{(i+1)}(\mathcal{O}) = V^{(i)}(\mathcal{O}) - \ln H^{(i+1)}(\mathcal{O}).

Repeat this process until H(O)H(\mathcal{O}) is sufficiently flat in the target interval, e.g., when the ratio of the maximum to the minimum is less than 2.

(c) Production run: Fix the converged V(O)V(\mathcal{O}) and perform large-scale sampling. Finally, recover the canonical ensemble expectation values by reweighting:

⟨O⟩β=∑iOieV(Oi)∑ieV(Oi),\langle\mathcal{O} \rangle_{\beta} = \frac{\sum_i \mathcal{O}_i e^{V(\mathcal{O}_i)}}{\sum_i e^{V(\mathcal{O}_i)}},

where O\mathcal{O} is the physical observation.

Multicanonical Hamiltonian Monte Carlo update

HMC uses Hamiltonian dynamics to evolve configurations, effectively reducing autocorrelation times, and has molecular-dynamics trajectories that parallelize well, making it more suitable for large-scale lattice simulations. By specializing the multicanonical framework of Appendix C to the case where an arbitrary differentiable order parameter O[ϕ]\mathcal{O}[\phi] is used as the driving variable, one can obtain a multicanonical Hamiltonian Monte Carlo method suitable for lattice field theory and statistical mechanics systems.

Adding the multicanonical effective action S^[ϕ]=βS[ϕ]+V(O[ϕ])\hat{S}[\phi] = \beta S[\phi] + V(\mathcal{O}[\phi]) to the HMC framework yields an order-parameter-driven multicanonical Hamiltonian Monte Carlo method. Introduce a momentum pp conjugate to the field variable, and define the guiding Hamiltonian,

HMD=12∑p2+S^[ϕ]=12∑p2+βS[ϕ]+V(O[ϕ]).H_{\mathrm{MD}} = \frac{1}{2}\sum p^2 + \hat{S}[\phi] = \frac{1}{2}\sum p^2 + \beta S[\phi] + V(\mathcal{O}[\phi]).

The accept/reject step uses the same Hamiltonian Hacc=HMDH_{\mathrm{acc}} = H_{\mathrm{MD}}. Differentiating with respect to the field variable gives the force

F(x)=−∂S^∂ϕ(x)=−β∂S∂ϕ(x)−dVdO∣O[ϕ]∂O∂ϕ(x).F(x) = -\frac{\partial\hat{S}}{\partial\phi(x)} = -\beta\frac{\partial S}{\partial\phi(x)} - \left.\frac{dV}{d\mathcal{O}}\right|_{\mathcal{O}[\phi]}\frac{\partial\mathcal{O}}{\partial\phi(x)}.

The first term −β ∂S/∂ϕ(x)-\beta\,\partial S/\partial\phi(x) is the standard physical force; the second term is the order-parameter biasing force, whose strength is controlled by the global scalar dV/dOdV/d\mathcal{O}, and whose direction is given by ∂O/∂ϕ(x)\partial\mathcal{O}/\partial\phi(x). For a large class of local or semi-local order parameters, e.g. O=1V∑xν(ϕ(x⃗)−A)∣ϕ(x⃗)−A∣\mathcal{O} = \frac{1}{\mathcal{V}}\sum_x^\nu(\phi(\vec{x})-A)|\phi(\vec{x})-A|, one has

∂O∂ϕ(x)=2V∣ϕ(x)−A∣.\frac{\partial\mathcal{O}}{\partial\phi(x)} = \frac{2}{\mathcal{V}}|\phi(x)-A|.

Therefore, the additional computational overhead of the biasing force is very small. Even for more complex order parameters, as long as their functional derivatives can be computed efficiently, the method remains practical.

The molecular dynamics integration uses a reversible, symplectic leapfrog scheme. Let the step size be Δt\Delta t, and the total number of steps be NMDN_{\mathrm{MD}}. Each step requires: computing the global order parameter O\mathcal{O} from the current configuration; obtaining dV/dOdV/d\mathcal{O} by table lookup or interpolation; computing the total force F(x)F(x); and updating the momenta and fields. A typical update sequence is

p(t+Δt/2)=p(t−Δt/2)−Δt F(ϕ(t)),p(t+\Delta t/2) = p(t-\Delta t/2) - \Delta t\,F(\phi(t)),
ϕ(t+Δt)=ϕ(t)+Δt p(t+Δt/2),\phi(t+\Delta t) = \phi(t) + \Delta t\,p(t+\Delta t/2),

and half-step corrections are added at the beginning and end of the trajectory to ensure time reversibility. After the trajectory ends, compute the change in the Hamiltonian,

ΔH=S^[ϕ′]+12∑p′2−(S^[ϕ]+12∑p2),\Delta H = \hat{S}[\phi'] + \frac{1}{2}\sum p'^2 - \left(\hat{S}[\phi] + \frac{1}{2}\sum p^2\right),

and accept the new configuration with probability

Pacc=min⁡(1,e−ΔH).P_{\mathrm{acc}} = \min\left(1,e^{-\Delta H}\right).

This guarantees that the sampling distribution is exactly e−S~e^{-\tilde{S}}, thereby obtaining samples with a flat order parameter. Since the MCHMC sampling is driven by the multicanonical potential, it has good parallel scalability and high ergodic efficiency.

Lattice-continuum matching and continuum extrapolation

This appendix collects the lattice-continuum matching and the derivation of the final lattice action used in the main text.

High-temperature effective potential and dimensional reduction

We start from the four-dimensional high-temperature effective potential for the electroweak phase transition,

V4(ϕ4)=12m42(T)ϕ42+16g4(T)ϕ43+124λ4ϕ44,(54)V_4(\phi_4) = \frac{1}{2}m_4^2(T)\phi_4^2 + \frac{1}{6}g_4(T)\phi_4^3 + \frac{1}{24}\lambda_4\phi_4^4, \tag*{(54)}

with

m42(T)=γ(T2−T02),g4(T)=−2αT,λ4=6λ.(55)m_4^2(T) = \gamma(T^2-T_0^2), \qquad g_4(T) = -2\alpha T, \qquad\lambda_4 = 6\lambda. \tag*{(55)}

where ϕ4\phi_4 is the scalar field, and γ\gamma, T0T_0, α\alpha, and λ\lambda are model dimensionless parameters. The potential is obtained from a high-temperature expansion and is valid when TT is large compared with the field-dependent mass scales.

At high temperature, the equilibrium thermodynamics is dominated by the zero Matsubara mode. The four-dimensional Euclidean action is

S4[ϕ4]=∫0βdτ∫d3x[12(∂τϕ4)2+12(∂iϕ4)2+V4(ϕ4)],(56)S_4[\phi_4] = \int_0^\beta d\tau\int d^3x \left[\frac{1}{2}(\partial_\tau\phi_4)^2 + \frac{1}{2}(\partial_i\phi_4)^2 + V_4(\phi_4)\right], \tag*{(56)}

with β=1/T\beta= 1/T. Expanding ϕ4\phi_4 in Matsubara modes, we get

ϕ4(τ,x)=ϕ0(x)+∑n≠0ϕn(x)eiωnτ,ωn=2πnT.(57)\phi_4(\tau,\mathbf{x}) = \phi_0(\mathbf{x}) + \sum_{n\ne0}\phi_n(\mathbf{x})e^{i\omega_n\tau}, \qquad\omega_n = 2\pi nT. \tag*{(57)}

With only the zero mode ϕ0(x)\phi_0(\mathbf{x}) kept, the time derivative vanishes and the time integral gives β=1/T\beta= 1/T, so

S4[ϕ4]≈1T∫d3x[12(∂iϕ0)2+V4(ϕ0)].(58)S_4[\phi_4] \approx\frac{1}{T}\int d^3x \left[\frac{1}{2}(\partial_i\phi_0)^2 + V_4(\phi_0)\right]. \tag*{(58)}

Defining the three-dimensional field

ϕ3=ϕ0T,(59)\phi_3 = \frac{\phi_0}{\sqrt{T}}, \tag*{(59)}

and the three-dimensional effective potential

V3(ϕ3)=1TV4(Tϕ3,T),(60)V_3(\phi_3) = \frac{1}{T}V_4(\sqrt{T}\phi_3,T), \tag*{(60)}

we obtain the three-dimensional action

S3[ϕ3]=∫d3x[12(∂iϕ3)2+V3(ϕ3)],(61)S_3[\phi_3] = \int d^3x \left[\frac{1}{2}(\partial_i\phi_3)^2 + V_3(\phi_3)\right], \tag*{(61)}

with S4≈S3S_4 \approx S_3 and the potential (D1):

V3(ϕ3)=12m3,phs2ϕ32+16g3,phsϕ33+124λ3,phsϕ34=12γ(T2−T02)ϕ32−13αT3/2ϕ33+14λTϕ34.(62)\begin{aligned} V_3(\phi_3) &= \frac{1}{2}m_{3,\mathrm{phs}}^2\phi_3^2 + \frac{1}{6}g_{3,\mathrm{phs}}\phi_3^3 + \frac{1}{24}\lambda_{3,\mathrm{phs}}\phi_3^4 \\ &= \frac{1}{2}\gamma(T^2-T_0^2)\phi_3^2 - \frac{1}{3}\alpha T^{3/2}\phi_3^3 + \frac{1}{4}\lambda T\phi_3^4. \tag*{(62)} \end{aligned}

This leading-order zero-mode reduction is self-consistent for the class of high-temperature potentials used here, which already incorporate the relevant thermal corrections. Therefore, no further TT-dependent matching is required beyond the zero-mode approximation[85].

To make the lattice simulation dimensionless, we introduce a scale M=150M = 150 GeV and rescale the field and couplings as

T30=T0M−1,T3=TM−1,T_3^0 = T_0M^{-1}, \qquad T_3 = TM^{-1},
ϕ=M−1/2ϕ3,λ3=M−1λ3,phs,\phi= M^{-1/2}\phi_3, \qquad\lambda_3 = M^{-1}\lambda_{3,\mathrm{phs}},
m32=M−2m3,phs2,g3=M−3/2g3,phs.(63)m_3^2 = M^{-2}m_{3,\mathrm{phs}}^2, \qquad g_3 = M^{-3/2}g_{3,\mathrm{phs}}. \tag*{(63)}

This rescaling leaves the action invariant but brings the field and couplings into dimensionless lattice units.

Field shift and elimination of the cubic term

The three-dimensional potential (D9) contains a cubic term. For the real scalar theory, it is convenient to shift the field by a constant,

ϕ=ψ+c,c=−g3(T)λ3(T).(64)\phi= \psi+ c, \qquad c = -\frac{g_3(T)}{\lambda_3(T)}. \tag*{(64)}

For fixed TT, both g3(T)g_3(T) and λ3(T)\lambda_3(T) are constants, so cc is constant. After the shift, the potential becomes

V3(ψ+c)=σ′ψ+12m′2(T)ψ2+124λ3(T)ψ4+constant,(65)V_3(\psi+c) = \sigma'\psi+ \frac{1}{2}m'^2(T)\psi^2 + \frac{1}{24}\lambda_3(T)\psi^4 + \text{constant}, \tag*{(65)}

with

σ′=−m32(T)g3(T)λ3(T)+g3(T)33λ3(T)2,(66)\sigma' = -\frac{m_3^2(T)g_3(T)}{\lambda_3(T)} + \frac{g_3(T)^3}{3\lambda_3(T)^2}, \tag*{(66)}
m′2(T)=m32(T)−g3(T)22λ3(T).(67)m'^2(T) = m_3^2(T) - \frac{g_3(T)^2}{2\lambda_3(T)}. \tag*{(67)}

The quartic coupling λ3(T)\lambda_3(T) is unchanged.

The reason for this shift is the following. In the shifted basis, the cubic coupling vanishes, glat=0g_{\mathrm{lat}}=0. Consequently, the coefficients of odd powers of the field are not renormalised, i.e.

δσ3=0,δg3=0.(68)\delta\sigma_{3}=0,\qquad\delta g_{3}=0. \tag*{(68)}

This eliminates the mixing between odd and even operators and simplifies the lattice-continuum matching considerably. One can then extract the lattice-continuum relations for the remaining even couplings from Refs. [78, 86, 87]. Although m′2(T)m^{\prime2}(T) and σ′\sigma^{\prime} still depend on TT, they are constant in a lattice simulation at fixed TT. Below we drop the primes and denote the shifted parameters simply by σ\sigma, m2m^{2}, and λ\lambda.

Matching relations

The bare lattice parameters are related to the MS parameters by

λlat=λ+δλ,mlat2=m2+δm2,σlat=σ,(69)\lambda_{\mathrm{lat}}=\lambda+\delta\lambda,\qquad m_{\mathrm{lat}}^{2}=m^{2}+\delta m^{2},\qquad \sigma_{\mathrm{lat}}=\sigma, \tag*{(69)}

together with the multiplicative factors ZϕZ_{\phi} and ZmZ_{m}. The counterterms δλ\delta\lambda, δm2\delta m^{2}, ZϕZ_{\phi}, and ZmZ_{m} are computed in lattice perturbation theory by matching zero-momentum correlation functions: the two- and four-point functions, the leading momentum dependence of the two-point function, and the insertion of ⟨ϕ2⟩\langle\phi^{2}\rangle. The couplings in three dimensions have positive mass dimension. Each additional loop order reduces the degree of ultraviolet divergence by one, so the matching can be organised order by order [78, 88, 89].

For the improved Laplacian (D21), the matching relations through O(a2)O(a^{2}) are [78, 79]

δλ=3ξλ2a2(4π)+λ3a2(4π)2(34ξ2−3C1−C23),\delta\lambda= \frac{3\xi\lambda^{2}a}{2(4\pi)} +\frac{\lambda^{3}a^{2}}{(4\pi)^{2}} \left(\frac{3}{4}\xi^{2}-3C_{1}-\frac{C_{2}}{3}\right),
Zϕ=1+C2λ2a26(4π)2,Z_{\phi}=1+\frac{C_{2}\lambda^{2}a^{2}}{6(4\pi)^{2}},
Zm=1+ξλa2(4π)+λ2a2(4π)2(ξ24−C12−C26),Z_{m}=1+\frac{\xi\lambda a}{2(4\pi)} +\frac{\lambda^{2}a^{2}}{(4\pi)^{2}} \left(\frac{\xi^{2}}{4}-\frac{C_{1}}{2}-\frac{C_{2}}{6}\right),
δm2=−Σλ2(4π)a+Σλ26(4π)2(ln⁡6aμ+C3−3Σξ).(70)\delta m^{2}=-\frac{\Sigma\lambda}{2(4\pi)a} +\frac{\Sigma\lambda^{2}}{6(4\pi)^{2}} \left(\ln\frac{6}{a\mu}+C_{3}-3\Sigma\xi\right). \tag*{(70)}

Here μ\mu is the MS renormalisation scale. The numerical constants for the improved Laplacian are [78]

Σ=2.75238391130752,\Sigma=2.75238391130752,
ξ=−0.083647053040968,\xi=-0.083647053040968,
C1=0.0550612,C_{1}=0.0550612,
C2=0.0334416,C_{2}=0.0334416,
C3=−0.86147916.(71)C_{3}=-0.86147916. \tag*{(71)}

These relations are the N=1N=1 Arnold–Moore results and are accurate up to O(a3)O(a^{3}) [86, 87], except for the mass parameter, which still has an O(a)O(a) residual correction. The operator ⟨ϕ2⟩\langle\phi^{2}\rangle also requires additive and multiplicative renormalisation [78]:

a⟨ϕ2⟩cont=Zm⟨ϕ2⟩lat−δ⟨ϕ2⟩.(72)a\langle\phi^{2}\rangle_{\mathrm{cont}} =Z_{m}\langle\phi^{2}\rangle_{\mathrm{lat}}-\delta\langle\phi^{2}\rangle. \tag*{(72)}

Final lattice action, improved Laplacian and dimensionless rescaling

After the above matching and shift, the lattice action takes the form

Slat=∑x[−12Zϕϕx(∇lat2ϕ)x+σlatϕx+12ZϕZmmlat2ϕx2+14!Zϕ2λlatϕx4].(73)S_{\mathrm{lat}}=\sum_{x}\left[ -\frac{1}{2}Z_{\phi}\phi_{x}(\nabla_{\mathrm{lat}}^{2}\phi)_{x} +\sigma_{\mathrm{lat}}\phi_{x} +\frac{1}{2}Z_{\phi}Z_{m}m_{\mathrm{lat}}^{2}\phi_{x}^{2} +\frac{1}{4!}Z_{\phi}^{2}\lambda_{\mathrm{lat}}\phi_{x}^{4} \right]. \tag*{(73)}

where ZϕZ_{\phi} and ZmZ_{m} are multiplicative renormalisation constants, and σlat\sigma_{\mathrm{lat}} is the coefficient of the linear term after the shift. The counterterms δλ\delta\lambda and δm2\delta m^{2} are included in the definitions of λlat\lambda_{\mathrm{lat}} and mlat2m_{\mathrm{lat}}^{2} through the matching relations in Appendix D 3. These counterterms are necessary to cancel lattice discretisation errors and to ensure that the continuum limit reproduces the correct MS physics. For the kinetic term, we use the O(a4)O(a^{4})-improved Laplacian

(∇lat2ϕ)x=−152ϕx+13∑i(ϕx+i+ϕx−i)−112∑i(ϕx+2i+ϕx−2i).(74)(\nabla_{\mathrm{lat}}^{2}\phi)_{x} =-\frac{15}{2}\phi_{x} +\frac{1}{3}\sum_{i}(\phi_{x+ i}+\phi_{x-i}) -\frac{1}{12}\sum_{i}(\phi_{x+2i}+\phi_{x-2i}). \tag*{(74)}

whose momentum-space form is

k^2=∑i(52−83cos⁡ki+16cos⁡2ki).(75)\hat{k}^{2}=\sum_{i}\left(\frac{5}{2}-\frac{8}{3}\cos k_{i}+\frac{1}{6}\cos2k_{i}\right). \tag*{(75)}

This improved Laplacian removes the leading O(a2)O(a^{2}) kinetic errors and is also required if the matching is carried to two loops. The lattice-continuum matching relations and the explicit expressions for ZϕZ_{\phi}, ZmZ_{m}, δλ\delta\lambda, and δm2\delta m^{2} are collected in Appendix D 3.

Overcooling and lattice implementation

The bare lattice mass parameter is related to the continuum MS mass by

mlat2=m2+δm2(a),(76)m_{\mathrm{lat}}^{2}=m^{2}+\delta m^{2}(a), \tag*{(76)}

where δm2(a)\delta m^{2}(a) is computed in lattice perturbation theory (see Eq. (70)), which is independent of m2m^{2} and depends only on λ3\lambda_{3} and aa [78]. Consequently, although the matching is performed through O(a3)O(a^{3}), the mass parameter still receives a residual correction starting at O(a)O(a), of the form

c3λ33a+c4λ34a2.(77)c_{3}\lambda_{3}^{3}a+c_{4}\lambda_{3}^{4}a^{2}. \tag*{(77)}

with c3c_3, c4c_4 pure numbers. These mass-independent lattice artifacts would enter physical observables and make the leading continuum extrapolation O(a)O(a) or O(a2)O(a^2) instead of the desired O(a3)O(a^3). However, since they are independent of the mass, they cancel exactly in the lattice difference

Δm′2≡m′2(T)−mc′2=mlat′2(a)−mc,lat′2(a),(78)\begin{aligned} \Delta m^{\prime2} &\equiv m^{\prime2}(T)-m_c^{\prime2}\\ &=m_{\mathrm{lat}}^{\prime2}(a)-m_{c,\mathrm{lat}}^{\prime2}(a), \tag*{(78)} \end{aligned}

which is also known as “overcooling”, where mc′m'_c is the critical shifted mass.

In the shifted potential (D12), where the cubic term is eliminated, we have

σ′=−m32g3λ3+g333λ32,m′2(T)=m32(T)−g3(T)22λ3(T).(79)\sigma'=-\frac{m_3^2g_3}{\lambda_3}+\frac{g_3^3}{3\lambda_3^2},\qquad m^{\prime2}(T)=m_3^2(T)-\frac{g_3(T)^2}{2\lambda_3(T)}. \tag*{(79)}

The critical temperature is determined by the exact Z2\mathbb{Z}_2-protected condition

σ′(Tc)=0,(80)\sigma'(T_c)=0, \tag*{(80)}

which is not corrected at any order in the loop expansion [80]. This condition is equivalent to

m32(Tc)=g3(Tc)23λ3(Tc).(81)m_3^2(T_c)=\frac{g_3(T_c)^2}{3\lambda_3(T_c)}. \tag*{(81)}

Using the temperature dependence

m32(T)=γ(T2−T02),g3(T)=−2αT3/2,λ3(T)=6λT,(82)m_3^2(T)=\gamma(T^2-T_0^2),\qquad g_3(T)=-2\alpha T^{3/2},\qquad\lambda_3(T)=6\lambda T, \tag*{(82)}

Eq. (D27) gives

Tc2=γT024α2(γ−18λ1)(83)T_c^2=\frac{\gamma T_0^2}{4\alpha^2\left(\gamma-\frac{18\lambda}{1}\right)} \tag*{(83)}

The critical shifted mass is then

mc′2=m3,c2−g3(Tc)22λ3(Tc)=−g3(Tc)26λ3(Tc).(84)m_c^{\prime2}=m_{3,c}^2-\frac{g_3(T_c)^2}{2\lambda_3(T_c)}=-\frac{g_3(T_c)^2}{6\lambda_3(T_c)}. \tag*{(84)}

The overcooling is defined as

Δm′2≡m′2(T)−mc′2,(85)\Delta m^{\prime2}\equiv m^{\prime2}(T)-m_c^{\prime2}, \tag*{(85)}

see Eq. (D25).

In the lattice simulation, for each lattice spacing aa we compute the critical bare mass

mc,lat′2(a)=mc′2+δm′2(a),(86)m_{c,\mathrm{lat}}^{\prime2}(a)=m_c^{\prime2}+\delta m^{\prime2}(a), \tag*{(86)}

and set the simulation bare mass to

mlat′2(a)=mc,lat′2(a)+Δm′2.(87)m_{\mathrm{lat}}^{\prime2}(a)=m_{c,\mathrm{lat}}^{\prime2}(a)+\Delta m^{\prime2}. \tag*{(87)}

Thus the mass-independent lattice corrections cancel exactly, and the physical overcooling is independent of aa. This reduces the leading lattice-spacing dependence of physical observables to O(a3)O(a^3), making the continuum extrapolation faster and more reliable.

For the parameters used in this work, we obtain

Tc≃1.78294,mc′2≃−1.48549,m′2(T)≃−1.26313,(88)T_c\simeq1.78294,\qquad m_c^{\prime2}\simeq-1.48549,\qquad m^{\prime2}(T)\simeq-1.26313, \tag*{(88)}

and therefore

Δm′2≃0.22236.(89)\Delta m^{\prime2}\simeq0.22236. \tag*{(89)}

In practice, for each aa we set mlat′2(a)=mc,lat′2(a)+0.22236m_{\mathrm{lat}}^{\prime2}(a)=m_{c,\mathrm{lat}}^{\prime2}(a)+0.22236.

References

References

  1. [1]V. A. Kuzmin, V. A. Rubakov, and M. E. Shaposhnikov, On the anomalous electroweak baryon number nonconservation in the early universe, Phys. Lett. B 155, 36 (1985).
  2. [2]V. A. Rubakov and M. E. Shaposhnikov, Electroweak baryon number non-conservation in the early universe and in high-energy collisions, Phys. Usp. 39, 461 (1996), arXiv:hep-ph/9603208 [hep-ph].DOI
  3. [3]J. M. Cline, A. Friedlander, D.-M. He, K. Kainulainen, B. Laurent, and D. Tucker-Smith, Baryogenesis and gravity waves from a UV-completed electroweak phase transition, Phys. Rev. D 103, 123529 (2021), arXiv:2102.12490 [hep-ph].DOI
  4. [4]J. M. Ellis, M. Lewicki, M. Merchand, J. M. No, and M. Zych, The scalar singlet extension of the Standard Model: gravitational waves versus baryogenesis, JHEP 01, 093, arXiv:2110.16305 [hep-ph].DOI
  5. [5]S. Bruggisser, B. von Harling, O. Matsedonskyi, and G. Servant, Status of electroweak baryogenesis in minimal composite Higgs, JHEP 08, 012, arXiv:2212.11953 [hep-ph].arxiv.org/abs/2212.11953
  6. [6]S. Kanemura and Y. Mura, Electroweak baryogenesis via top-charm mixing, JHEP 09, 153, arXiv:2303.11252 [hep-ph].arxiv.org/abs/2303.11252
  7. [7]C. Idegawa and E. Senaha, Electron electric dipole moment and electroweak baryogenesis in a complex singlet extension of the Standard Model with degenerate scalars, Phys. Lett. B 848, 138332 (2024), arXiv:2309.09430 [hep-ph].arxiv.org/abs/2309.09430
  8. [8]C. J. Hogan, Magnetohydrodynamic Effects of a First-Order Cosmological Phase Transition, Phys. Rev. Lett. 51, 1488 (1983).DOI
  9. [9]Y. Di, J. Wang, R. Zhou, L. Bian, R.-G. Cai, and J. Liu, Magnetic Field and Gravitational Waves from the First-Order Phase Transition, Phys. Rev. Lett. 126, 251102 (2021), arXiv:2012.15625 [astro-ph.CO].DOI
  10. [10]K. Enqvist and P. Olesen, On primordial magnetic fields of electroweak origin, Phys. Lett. B 319, 178 (1993), arXiv:hep-ph/9308270.DOI
  11. [11]J. Yang and L. Bian, Magnetic field generation from bub- ble collisions during first-order phase transition, Phys. Rev. D 106, 023510 (2022), arXiv:2102.01398 [astro-ph.CO].
  12. [12]T. Vachaspati, Magnetic fields from cosmological phase transitions, Phys. Lett. B 265, 258 (1991).DOI
  13. [13]Y. Di, L. Bian, and R.-G. Cai, Baryogenesis Induced by Magnetic Field Effects During the Electroweak Phase Transition, (2024), arXiv:2409.16124 [hep-ph].arxiv.org/abs/2409.16124
  14. [14]Y. Di, L. Bian, and R.-G. Cai, Impact of primordial magnetic fields on the first-order electroweak phase transition, Phys. Rev. D 113, 043529 (2026), arXiv:2508.07416 [hep-ph].arxiv.org/abs/2508.07416
  15. [15]G. Baym, D. Bodeker, and L. D. McLerran, Magnetic fields produced by phase transition bubbles in the electroweak phase transition, Phys. Rev. D 53, 662 (1996), arXiv:hep-ph/9507429.DOI
  16. [16]D. Grasso and A. Riotto, On the nature of the magnetic fields generated during the electroweak phase transition, Phys. Lett. B 418, 258 (1998), arXiv:hep-ph/9707265.DOI
  17. [17]J. M. Quashnock, A. Loeb, and D. N. Spergel, Magnetic Field Generation During the Cosmological QCD Phase Transition, Astrophys. J. Lett. 344, L49 (1989).DOI
  18. [18]J. Liu, L. Bian, R.-G. Cai, Z.-K. Guo, and S.-J. Wang, Primordial black hole production during first-order phase transitions, Phys. Rev. D 105, L021303 (2022), arXiv:2106.05637 [astro-ph.CO].DOI
  19. [19]G. Franciolini, Y. Gouttenoire, and R. Jinno, Curvature Perturbations from First-Order Phase Transitions: Implications to Black Holes and Gravitational Waves, Phys. Rev. Lett. 136, 171404 (2026), arXiv:2503.01962 [hep-ph].
  20. [20]B. Carr, A. J. Iovino, G. Perna, V. Vaskonen, and H. Veermäe, Primordial black holes: constraints, potential evidence and prospects, Riv. Nuovo Cim. 49, 225 (2026), arXiv:2601.06024 [astro-ph.CO].DOI
  21. [21]M. Lewicki, P. Toczek, and V. Vaskonen, Primordial black holes from strong first-order phase transitions, JHEP 09, 092, arXiv:2305.04924 [astro-ph.CO].DOI
  22. [22]Y. Gouttenoire and T. Volansky, Primordial black holes from supercooled phase transitions, Phys. Rev. D 110, 043514 (2024), arXiv:2305.04942 [hep-ph].DOI
  23. [23]K. Jedamzik and J. C. Niemeyer, Primordial black hole formation during first order phase transitions, Phys. Rev. D 59, 124014 (1999), arXiv:astro-ph/9901293.DOI
  24. [24]M. B. Hindmarsh, M. Luben, J. Lumma, and M. Pauly, Phase transitions in the early universe, SciPost Phys. Lect. Notes 24, 1 (2021), arXiv:2008.09136 [astro-ph.CO].DOI
  25. [25]C. Caprini et al., Science with the space-based interferometer eLISA. II: Gravitational waves from cosmological phase transitions, JCAP 04, 001, arXiv:1512.06239 [astro-ph.CO].DOI
  26. [26]C. Caprini et al., Detecting gravitational waves from cosmological phase transitions with LISA: an update, JCAP 03, 024, arXiv:1910.13125 [astro-ph.CO].DOI
  27. [27]P. Auclair et al. (LISA Cosmology Working Group), Cosmology with the Laser Interferometer Space Antenna, Living Rev. Rel. 26, 5 (2023), arXiv:2204.05434 [astro-ph.CO].DOI
  28. [28]G. Agazie et al. (NANOGrav), The NANOGrav 15 yr Data Set: Evidence for a Gravitational-wave Background, Astrophys. J. Lett. 951, L8 (2023), arXiv:2306.16213 [astro-ph.HE].DOI
  29. [29][29*] J. Antoniadis et al. (EPTA, InPTA), The second data release from the European Pulsar Timing Array - III. Search for gravitational wave signals, Astron. Astrophys. 678, A50 (2023), arXiv:2306.16214 [astro-ph.HE].
  30. [30]H. Xu et al., Searching for the nano-Hertz stochastic gravitational wave background with the Chinese Pulsar Timing Array data release I, Res. Astron. Astrophys. 23, 075024 (2023), arXiv:2306.16216 [astro-ph.HE].DOI
  31. [31]D. J. Reardon et al., Search for an Isotropic Gravitational-wave Background with the Parkes Pulsar Timing Array, Astrophys. J. Lett. 951, L6 (2023), arXiv:2306.16215 [astro-ph.HE].DOI
  32. [32]L. Bian, R.-G. Cai, J. Liu, X.-Y. Yang, and R. Zhou, Evidence for different gravitational-wave sources in the NANOGrav dataset, Phys. Rev. D 103, L081301 (2021), arXiv:2009.13893 [astro-ph.CO].
  33. [33]L. Bian, S. Ge, J. Shu, B. Wang, X.-Y. Yang, and J. Zong, Gravitational wave sources for pulsar timing arrays, Phys. Rev. D 109, L101301 (2024), arXiv:2307.02376 [astro-ph.HE].arxiv.org/abs/2307.02376
  34. [34]I. Y. Kozbarev, L. B. Okun, and M. B. Voloshin, Bubbles in metastable vacuum, Yad. Fiz. 20, 1229 (1974).
  35. [35]S. R. Coleman, The Fate of the False Vacuum. 1. Semiclassical Theory, Phys. Rev. D 15, 2929 (1977), [Erratum: Phys. Rev. D 16, 1248 (1977)].
  36. [36]C. G. Callan, Jr. and S. R. Coleman, The Fate of the False Vacuum. 2. First Quantum Corrections, Phys. Rev. D 16, 1762 (1977).DOI
  37. [37]J. S. Langer, Theory of the condensation point, Annals Phys. 41, 108 (1967).
  38. [38]J. S. Langer, Statistical theory of the decay of metastable states, Annals Phys. 54, 258 (1969).DOI
  39. [39]J. S. Langer, Metastable states, Physica 73, 61 (1974).DOI
  40. [40]J. W. Cahn and J. E. Hilliard, Free Energy of a Nonuniform System. III. Nucleation in a Two-Component Incompressible Fluid, J. Chem. Phys. 31, 688 (1959).
  41. [41]A. D. Linde, Fate of the False Vacuum at Finite Temperature: Theory and Applications, Phys. Lett. B 100, 37 (1981).DOI
  42. [42]I. Armeck, Quantum Statistical Metastability, Phys. Rev. Lett. 46, 388 (1981).
  43. [43]K. Kajantie, M. Laine, K. Rummukainen, and M. E. Shaposhnikov, Generic rules for high temperature dimensional reduction and their application to the standard model, Nucl. Phys. B 458, 90 (1996), arXiv:hep-ph/9508379 [hep-ph].DOI
  44. [44]E. Braaten and A. Nieto, Effective field theory approach to high temperature thermodynamics, Phys. Rev. D 51, 6990 (1995), arXiv:hep-ph/9501375 [hep-ph].DOI
  45. [45]K. K. Kainulainen, V. Keus, L. Niemi, K. Rummukainen, T. V. I. Tenkanen, and V. Vaskonen, On the validity of perturbative studies of the electroweak phase transition in the Two Higgs Doublet model, JHEP 06, 075, arXiv:1904.01329 [hep-ph].
  46. [46]O. Gould, J. Kozaczuk, L. Niemi, M. J. Ramsey-Musolf, T. V. I. Tenkanen, and D. J. Weir, Nonperturbative analysis of the gravitational waves from a first-order electroweak phase transition, Phys. Rev. D 100, 115024 (2019), arXiv:1903.11604 [hep-ph].DOI
  47. [47]A. Ekstedt and J. Löfgren, A critical look at the electroweak phase transition, JHEP 12, 136, arXiv:2006.12614 [hep-ph].
  48. [48][48*] D. Croon, O. Gould, P. Schicho, T. V. I. Tenkanen, and G. White, Theoretical uncertainties for cosmological first-order phase transitions, JHEP 04, 055, arXiv:2009.10080 [hep-ph].
  49. [49]O. Gould and T. V. I. Tenkanen, On the perturbative expansion at high temperature and implications for cosmological phase transitions, JHEP 06, 069, arXiv:2104.04399 [hep-ph].DOI
  50. [50]A. Ekstedt, Convergence of the nucleation rate for first-order phase transitions, Phys. Rev. D 106, 095026 (2022), arXiv:2205.05145 [hep-ph].DOI
  51. [51]D. J. H. Chung, A. J. Long, and L.-T. Wang, 125 GeV Higgs boson and electroweak phase transition model classes, Phys. Rev. D 87, 023509 (2013), arXiv:1209.1819 [hep-ph].DOI
  52. [52]A. Ekstedt, O. Gould, and J. Lofgren, Radiative first-order phase transitions to next-to-next-to-leading order, Phys. Rev. D 106, 036012 (2022), arXiv:2205.07241 [hep-ph].DOI
  53. [53]O. Gould and T. V. I. Tenkanen, Perturbative effective field theory expansions for cosmological phase transitions, JHEP 01, 048, arXiv:2309.01672 [hep-ph].arxiv.org/abs/2309.01672
  54. [54]J. Liu, R. Qin, and L. Bian, Towards accurate gravitational wave predictions: gauge-invariant nucleation in the electroweak phase transition, JHEP 10, 086, arXiv:2601.05793 [hep-ph].
  55. [55]J. Liu, R. Qin, and L. Bian, Gauge-invariant bubble nucleation and gravitational waves from first-order electroweak phase transitions, Phys. Rev. D 114, L031703 (2026), arXiv:2512.05565 [hep-ph].arxiv.org/abs/2512.05565
  56. [56]E. J. Weinberg and A.-q. Wu, Understanding complex perturbative effective potentials, Phys. Rev. D 36, 2474 (1987).pubmed.ncbi.nlm.nih.gov/9958456
  57. [57]C. L. Wainwright, CosmoTransitions: Computing cosmological phase transition temperatures and bubble profiles with multiplets, Comput. Phys. Commun. 183, 2006 (2012), arXiv:1109.4189 [hep-ph].DOI
  58. [58]C. Delaunay, C. Grojean, and J. D. Wells, Dynamics of non-renormalizable electroweak symmetry breaking, JHEP 04, 029, arXiv:0711.2511 [hep-ph].DOI
  59. [59]P. Athron, C. Balazs, A. Fowlie, L. Morris, G. White, and Y. Zhang, How arbitrary are perturbative calculations of the electroweak phase transition?, JHEP 01, 050, arXiv:2208.01319 [hep-ph].DOI
  60. [60]Y. Zhu, J. Liu, R. Qin, and L. Bian, Theoretical uncertainties in first-order electroweak phase transitions, Phys. Rev. D 112, 015018 (2025), arXiv:2503.19566 [hep-ph].arxiv.org/abs/2503.19566
  61. [61]D. Y. Grigoriev and V. A. Rubakov, Soliton pair creation at finite temperatures. Numerical study in (1+1)-dimensions, Nucl. Phys. B 299, 67 (1988).
  62. [62]O. T. Valls and G. F. Mazenko, Nucleation in a time-dependent Ginzburg-Landau model: A numerical study, Phys. Rev. B 42, 6614 (1990).pubmed.ncbi.nlm.nih.gov/9994748
  63. [63]M. G. Alford, H. Feldman, and M. Gleiser, Thermal nucleation of kink-antikink pairs, Phys. Rev. Lett. 68, 1645 (1992).pubmed.ncbi.nlm.nih.gov/10045184
  64. [64]M. G. Alford, H. Feldman, and M. Gleiser, Thermal activation of metastable decay: Testing nucleation theory, Phys. Rev. D 47, R2168 (1993).pubmed.ncbi.nlm.nih.gov/10015864
  65. [65]M. G. Alford and M. Gleiser, Metastability in two-dimensions and the effective potential, Phys. Rev. D 48, 2838 (1993), arXiv:hep-ph/9304245 [hep-ph].DOI
  66. [66]S. Borsanyi, A. Patkos, J. Polonyi, and Z. Szep, Fate of the classical false vacuum, Phys. Rev. D 62, 085013 (2000), arXiv:hep-th/0004059 [hep-th].DOI
  67. [68]D. Pirvu, M. C. Johnson, and S. Sibiryakov, Bubble velocities and oscillon precursors in first order phase transitions, JHEP 11, 064, arXiv:2312.13364 [hep-th].DOI
  68. [69]B. A. Berg and T. Neuhaus, Multicanonical algorithms for first order phase transitions, Phys. Lett. B 267, 249 (1991).DOI
  69. [70]B. A. Berg and T. Neuhaus, Multicanonical ensemble: A new approach to simulate first order phase transitions, Phys. Lett. 68, 9 (1992), arXiv:hep-lat/9202004 [hep-lat].DOI
  70. [71]G. D. Moore and K. Rummukainen, Electroweak bubble nucleation, nonperturbatively, Phys. Rev. D 63, 045002 (2001), arXiv:hep-ph/0009132 [hep-ph].DOI
  71. [72]K. Binder and M. H. Kalos, Critical clusters in a supersaturated vapor: Theory and Monte Carlo simulation, J. Stat. Phys. 22, 363 (1980).DOI
  72. [73]K. Binder, Theory of the evaporation/condensation transition of equilibrium droplets in finite volumes, Physica A 319, 99 (2003).DOI
  73. [74]J. Hällfors and K. Rummukainen, Expanded ensemble method for bubble nucleation, Phys. Rev. D 111, 114512 (2025).arxiv.org/abs/2502.14610
  74. [75]G. Arnold, K. Schilling, and T. Lippert, Multicanonical hybrid Monte Carlo algorithm: Boosting simulations of compact QED, Phys. Rev. D 59, 054509 (1999), arXiv:hep-lat/9809160 [hep-lat].DOI
  75. [76]S. Duane, A. D. Kennedy, B. J. Pendleton, and D. Roweth, Hybrid Monte Carlo, Phys. Lett. B 195, 216 (1987).DOI
  76. [77]T. Neuhaus and J. S. Hager, 2D crystal shapes, droplet condensation, and exponential slowing down in simulations of first-order phase transitions, J. Stat. Phys. 113, 47 (2003).DOI
  77. [78]G. D. Moore, K. Rummukainen, and A. Tranberg, Non-perturbative computation of the bubble nucleation rate in the cubic anisotropy model, JHEP 04, 017, arXiv:hep-lat/0103036 [hep-lat].DOI
  78. [79]O. Gould, A. Kormu, and D. J. Weir, A nonperturbative test of nucleation calculations for strong phase transitions, (2024), arXiv:2404.01876 [hep-th].DOI
  79. [80]O. Gould, Real scalar phase transitions: a nonperturbative analysis, JHEP 04, 057, arXiv:2102.07531 [hep-ph].
  80. [81]A. Ekstedt, O. Gould, and J. Hirvonen, Bubbledet: a python package to compute functional determinants for bubble nucleation, JHEP 12, 056, arXiv:2308.16552 [hep-ph].
  81. [82]O. Gould, S. Guyer, and K. Rummukainen, First-order electroweak phase transitions: A nonperturbative update, Phys. Rev. D 106, 114507 (2022), arXiv:2205.07238 [hep-lat].DOI
  82. [83]A. Berera, J. Mabillard, B. W. Mintz, and R. O. Ramos, Formulating the Kramers problem in field theory, Phys. Rev. D 100, 076005 (2019), arXiv:1906.08684 [hep-ph].DOI
  83. [84]H. A. Kramers, Brownian motion in a field of force and the diffusion model of chemical reactions, Physica 7, 284 (1940).DOI
  84. [85]K. Enqvist, J. Ignatius, K. Kajantie, and K. Rummukainen, Nucleation and bubble growth in a first-order cosmological electroweak phase transition, Phys. Rev. D 45, 3415 (1992).pubmed.ncbi.nlm.nih.gov/10014238
  85. [86]P. Arnold and G. D. Moore, Monte carlo simulation of O(2) ϕ⁴ field theory in three dimensions, Phys. Rev. E 64, 066113 (2001), arXiv:cond-mat/0103227 [cond-mat].arxiv.org/abs/cond-mat/0103227
  86. [87]X. Sun, Monte carlo studies of three-dimensional O(1) and O(4) φ⁴ theory related to Bose-Einstein condensation phase transition temperatures, Phys. Rev. E 67, 066702 (2003), arXiv:hep-lat/0209144 [hep-lat].DOI
  87. [88]M. Laine, Exact relation of lattice and continuum parameters in three-dimensional SU(2) + Higgs theories, Nucl. Phys B 451, 484 (1995), arXiv:hep-lat/9504001 [hep-lat].DOI
  88. [89]M. Laine and A. Rajantie, Lattice continuum relations for 3-D SU(N) + Higgs theories, Nucl. Phys. B 513, 471 (1998), arXiv:hep-lat/9705003 [hep-lat].DOI

Paper details

vlnc:W_4w9bz2gb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents